For a year, I've been sporadically writing blog posts exploring why Generative AI is poised to drive academic and business outcomes. In my previous posts, I focused on building trust, safety and transparency in Generative AI applications, and before that, wrote about career readiness preparation, and how organizations can leverage the unique properties of Large Language Models (LLMs) to provide personalized feedback for their end users, at scale, on a 24/7, 365, on-demand basis. In this blog, I will continue to build on that chain of thought (pun intended) and explore the world of continuous learning and development (L&D) in corporate settings. There are parts of this blog that may require a deeper dive into some technical bits, references to which I have provided in the blog. For those of you that want to skip past these bits, but still want to tap into the power of Generative AI for L&D, please feel free to reach out to me directly and I will do my best to guide you or help.

Optimizing your workforce: An important topic of discussion.
In today’s fast-paced corporate landscape, the ability to harness data effectively can make a world of difference in driving business outcomes. Among the tools available, conversational analytics, powered by Generative AI Models, emerges as a game-changer, providing insights that are often overlooked. By analyzing the nuances of employee interactions, organizations can uncover patterns, preferences, and areas for improvement, ultimately fostering a culture of continuous learning and adaptability. As companies strive to enhance engagement and productivity, leveraging the power of conversational analytics equips leaders with the knowledge necessary to tailor training initiatives and optimize workforce capabilities. This data-driven approach emphasizes an important aspect of talent development: the need to map learning and development (L&D) efforts to business outcomes, ensuring that skills acquired during training translate directly into enhanced organizational performance.

The Need to Map L&D to Business Outcomes

This article from the HR Magazine highlights research and several interesting statistics that show how hungry people are for opportunities to learn on the job. And this translates directly to a burden on the organization’s Learning & Development (L&D) departments, who are often short-staffed, resource constrained, or both. I have seen and experienced too often, the misconception that learning and development is primarily associated with content creation, and serving up tutorials, lessons, regulatory sessions, or similar modules, at discrete instants in time, to check a box within an organization. This highlights a pervasive misconception that learning and development (L&D) is primarily associated with content creation and compliance tasks done only at specific intervals. Often, there is the exercise of “needing to map” the learning and development function within the organization to the organization’s objectives or goals. While there are healthy debates and multiple viewpoints on this aspect of L&D (and there is no right or wrong approach, since all organizational cultures and workforces are nuanced and unique), there is also a seismic shift with Generative AI that is poised to flip the “need for mapping” on its head. By tapping into the underlying architecture of Large Language Models, we can do things that weren’t quite possible before. For example, by having a data strategy in place, Generative AI can help automatically identify and map the learning & development functions based on business outcomes. This is where we start to get into the interesting world of Predictive AI. If this topic is still holding your attention, read on, because I will try to explain this idea further with some real examples.
​
Applications for Sales & Manager Training
Predictive analytics offers organizations the ability to forecast training needs and outcomes by analyzing patterns from historical real-world data. The staggering statistic below that I computed using some back-of-the-envelope math should give anyone doubting if we have any available data, food for second thought:

So, what are organizations doing with all this data? More importantly, what can organizations do with this data, particularly if they were to begin using advanced Conversational Analytics? The possibilities are endless, but what matters most is making sure that the choice is well-aligned with business outcomes.

​As an example, in the realm of sales training, organizations can utilize predictive analytics to identify the characteristics of top performers and tailor programs that replicate their success traits. By understanding which training modules yield the best results, such as increased sales conversions or improved customer interactions, companies can refine their approaches to training and ensure maximum impact.

Similarly, for manager training, predictive analytics can help assess the current competencies of managers and forecast the skills required for future challenges. By examining employee feedback, performance assessments, and other data sources, organizations can identify gaps in managerial skills and create targeted development plans. This proactive approach ensures that training is not just reactive to current needs but anticipates future demands, ultimately equipping leaders with the tools necessary to drive team performance and business success.
And this brings me to the foundation for all of this to occur. It is imperative that organizations invest in a good data strategy that can support their workforce and L&D departments.

​The Data Strategy
Until now, we didn’t have the ability to look at vast volumes of data and comb through them to extract insights that mattered to our academic or business context. This also meant that we didn’t really have a data strategy in place that would directly impact our future business outcomes. Sure, we all collected and saved data that mattered to us, but this data was unstructured at best, and more likely, unusable at worst. Large Language Models (LLMs) have changed that. We now know exactly how data can be used to improve the performance of these models. We also know how we can alter the final layers of these incredible neural networks with our own data to achieve novel results. In a nutshell, this data can be broken up into the following categories:

• Pre-training corpora or foundational datasets used to train LLMs on large amounts of text, that allows them to perform next word prediction, grammar, patterns, or even (apparent) reasoning abilities.
• Datasets that are used to adapt the functionality of LLMs in order to perform very specific tasks such as classification or instruction following through the process of fine-tuning.
• Evaluation datasets to check whether the LLMs that were built from scratch or fine-tuned are performing to the standards that you require in your academic or business context.
• Ad-hoc datasets that serve as “training wheels”, when used in conjunction with traditional LLM usage e.g. prompt-engineering with one-shot or few-shot learning.

A comprehensive overview of these datasets for LLMs can be found here on arXiv. Also, if you’re interested in getting some hands-on experience with LLMs, I could not more highly recommend the book “Build a Large Language Model (From Scratch)” by Sebastian Raschka, and his associated GitHub repository.

The Connection between L&D, Data Strategy, and Business Outcomes
At this point, you’re probably thinking: "How does this help me with L&D and the need to map the function to business outcomes?" Well, from the previous section, we know that we can transform any data that we have into insights using the power of Large Language Models. So, the questions that we should be asking ourselves could be something along these lines, in the following order:

• What are the biggest challenges or problems that our business is facing?
• What kind of data (quality, quantity, and category from the previous section) do we have, that is related to those challenges?
• What other data can we be collecting that is directly related to those challenges that we’re facing?
• What kind of insights do we think we could gain, if we had an infinite number of resources to comb through that data?

If you haven’t already, you will soon be coming up with ideas in your mind on what you’d like to do with LLMs. I’d like to help kickstart that process by focusing on one of the most accumulated datasets that organizations will likely have access to: Recorded virtual meetings or calls, involving human conversation.

Applying Conversational Analytics for Insight Extraction
The potential of conversational analytics, powered by Generative AI, in corporate settings is vast, particularly when it comes to extracting actionable insights from recorded meetings and calls. By leveraging the transformer architecture of LLMs, organizations can analyze conversations to identify trends, sentiment, and key topics that are impacting business performance.

• Identifying Pain Points and Opportunities: Analyzing conversations allows organizations to pinpoint recurring themes related to customer feedback or employee issues. For instance, topics such as product defects or recurring customer service complaints can be automatically flagged for further investigation. This insight grants L&D teams the ability to directly address these concerns through tailored training programs.
• Enhancing Collaboration and Communication: By evaluating the dynamics of team discussions, companies can gain a better understanding of collaboration patterns. Are there communication gaps that need addressing? Do certain teams struggle with alignment on business goals? These insights can guide targeted training sessions aimed at improving teamwork and enhancing overall productivity.
• Feedback Loops for Continuous Improvement: Implementing conversational analytics creates a framework for continuous feedback and improvement. L&D initiatives can be adapted based on real-time insights gathered from employee discussions, ensuring that training remains relevant and effective in meeting evolving business needs.

​Measuring the Impact of Durable Skills on Business Outcomes
Once conversational analytics have been integrated into L&D strategies, measuring their impact, becomes essential for long-term business success. We can do this particularly well by leveraging the durable skills framework. Durable skills, alternatively referred to as "soft skills", such as communication, empathy, persuasion, negotiation, problem-solving, critical thinking, adaptability, and teamwork, play a critical role in enhancing employee performance and driving business outcomes. And they lend themselves very well for measurement in conjunction with conversational analytics.

To effectively evaluate success, I’d argue that organizations who are serious about positive change through training should consider the following aspects:

• Setting Clear KPIs for Durable Skills: Key Performance Indicators (KPIs) must be established to assess the effectiveness of training programs that focus on developing durable skills. I have seen one too many organizations shy away from this initial hurdle. Metrics such as employee collaboration levels, conflict resolution success rates, and customer interaction improvements need to be defined for a training program to work. For example, a surge in team cohesion after a collaboration-based training initiative can serve as a strong indicator of success.
• Linking Skill Development to Business Metrics: Organizations should actively connect the development of durable skills to broader business performance metrics. For instance, if a training intervention aimed at enhancing communication skills leads to improved sales conversations and higher closing rates, this serves as a clear proof point. Analyzing the correlation between training programs and these key outcomes demonstrates the value of focusing on durable skills.
• Utilizing Dashboards to Visualize Skill Impact: By creating comprehensive reporting tools that assess the growth of durable skills alongside business performance, stakeholders can gain insights into the effectiveness of L&D efforts. Dashboards that track progress in areas like communication and teamwork will help sustain executive buy-in and guide resource allocation for future skill development initiatives.

​Conclusion: The Future of L&D through Conversational Analytics
In conclusion, the integration of conversational analytics within Corporate Learning and Development strategies presents a profound opportunity for organizations to drive meaningful business outcomes. By harnessing the power of data derived from conversations, companies can gain insights that inform training initiatives, address key business challenges, and foster a culture of continuous improvement. As the workplace landscape continues to shift towards hybrid and remote settings, the relevance of conversational analytics will only increase. Organizations that leverage these insights will not only optimize their workforce but also remain agile in navigating the complexities of the modern business environment. The future of work lies in data-driven strategies where informed decision-making leads to impactful learning experiences, ultimately translating into enhanced performance and growth.
At Relativ, we help organizations experiment with their own data and understand how LLMs work with different contextual information, so they can expand these capabilities and begin to measure various skills that individuals exhibit during their conversational exchanges. We have developed advanced AI models that accurately measure essential durable skills relevant across multiple industries. The consistent performance of these models across diverse contexts reinforces the essence of “durable skills”, highlighting their ability to seamlessly transfer and apply across various work environments. These models enable us to create simulated environments that facilitate high-stakes conversations, where individuals can practice and refine their durable skills. Through personalized feedback, we empower users to enhance their performance in these critical interactions. To learn more, and get a demo of our proprietary conversational analytics engine, head over to our website or reach out to us to learn how we can help you deploy your own AI models, infused with psychology, and linguistics, to empower your organization and end users with the durable skills they require to meet the challenges of the future of work.

The last 18 months or so have been one of the most exciting periods for growth of AI. It’s interesting that while the field of AI has been around for decades and maturing rapidly on its own, the introduction of Large Language Models and Generative AI has created a massive amount of adoption and interest across all communities. Perhaps one of the easiest explanations for this is the nature of the output from these AI models. They cater to one of our social needs – the need for communication / use of language to exchange information.

Unlike traditional AI models, these textual or conversational outputs from Generative AI models are not mathematical in nature or require any additional explanation. Instead, they can deliver information to us in a form that we easily understand every day. More importantly, we can “instruct” these models to provide us with outputs in a manner that we can comprehend easily, again, without the need for sophisticated programming languages or the mathematical basis required previously. In my introductory blog post, I outline some of the characteristics of Large Language Models and why they are well-suited to drive academic or business outcomes. I also highlighted the true nature of such models, the black box problem, and emergent behavior, all of which essentially boil down to this: While the application of LLMs can have a profound impact in helping societies, how LLMs work internally remains an important area of research. 

As more such Generative AI models become commonplace, researchers are making progress in trying to understand “how” these Large Language Models interpret information and whether their outputs can be trusted. In one such recent such ground-breaking development, researchers at OpenAI attempted to decompose the internal representation of GPT-4 into millions of interpretable patterns. According to the researchers:

​In the next section, I will attempt to explain autoencoders, a type of artificial neural network, that can potentially provide some initial insights into the inner workings of large language models such as GPTs (Generative Pre-trained Transformers). Autoencoders were used by the researchers at OpenAI to identify activation patterns in the neurons and correlate them to identifiable concepts that were understandable by humans.   

Autoencoders
Autoencoders can help understand the inner workings of a neural network by providing insights into how data is represented and processed within the network. In simple terms, it compresses input data into a smaller feature set and then attempts to reconstruct the original input from this compressed set of features. IBM has an excellent article on autoencoders for those who want more details. Here’s how autoencoders can be useful in understanding models like GPT-4:

Dimensionality Reduction and Visualization:

• Latent Space Representation: The middle layer of an autoencoder (latent space) captures a compressed version of the input data. By analyzing these latent representations, we can understand the essential features that the network considers important.
• Visualization: Dimensionality reduction techniques (t-SNE, PCA) applied to the latent space can help visualize high-dimensional data in 2D or 3D. This can reveal clusters and patterns, showing how the network groups similar inputs together.

Feature Extraction

• Sparse Features: Sparse autoencoders can help identify a small number of key features that the network uses to make decisions. These features can be more interpretable than the raw input data.
• Interpretable Features: By examining which features are activated for different inputs, we can infer what concepts the network is recognizing and how it processes information.

Reconstruction Error Analysis

• Understanding Errors: By comparing the original input with the reconstructed output, we can identify where the autoencoder struggles to accurately recreate the input. This can highlight which aspects of the data are harder for the network to capture and may point to areas for improvement.
• Outliers and Anomalies: High reconstruction errors can indicate outliers or anomalies in the data, helping us understand the limits of the network’s generalization capabilities.

Layer-wise Inspection

• Intermediate Layers: By examining the outputs of the encoder’s intermediate layers, we can see how the data is transformed step-by-step. This can provide insights into how the network processes and abstracts information.
• Activation Patterns: Studying the activation patterns of different layers can reveal which parts of the network are most active for certain inputs, allowing us to understand the internal decision-making process a little better.

Simplifying Complex Models

• Model Compression: Autoencoders can be used to compress complex models into simpler representations. Understanding these simpler models can give us insights into the essential components and processes of the original complex model.
• Knowledge Distillation: The knowledge captured by an autoencoder in its latent space can be used to train simpler models, making it easier to analyze and interpret the learned features.

​Implementation & Results
For those of you interested in code and practical implementation of autoencoders, I recommend starting with some basics. Use keras and tensorflow to perform these steps:

• Load a dataset.
• Normalize and flatten the data.
• Define the compression dimensions.
• Define the layers for the encoder and the decoder.
• Create the autoencoder model to map the input to its reconstruction.
• Compile and Train the model.
• Encode and decode some test data and visualize the results.

Here is an example output where I trained an autoencoder using the above framework to reconstruct hand-drawn images. The results, as you can see, speak for themselves.

Once you understand how autoencoders work, we can look at how to apply this to better comprehend the inner workings of LLMs.

​Applications of Autoencoders to understanding LLMs.
As you are reading through this next section, please refer to the SAE Viewer from OpenAI to understand how activations occur for different concepts for GPT-4 or GPT-2. In the context of large language models (LLMs), autoencoders can be trained on the activations of the LLM. This helps to identify and visualize the features that the LLM uses to understand and generate language. To best understand this, it is worth looking at an example with some data that we can all relate to.

• I trained an autoencoder to load LLM embeddings using a pre-trained BERT model and tokenizer (head over to hugging face to explore this)
• I then provided it with samples of text of varying degree of “empathy”, a durable skill that I predict will matter a lot to the future workforce.
• I defined an autoencoder, and compiled and trained the model as previously described in the example with the hand-drawn images of numbers.
• I encoded the embeddings to reduce dimensionality using this autoencoder.
• I also directly reduced the dimensionality of the embeddings, knowing that I would not benefit from the non-linearity preservation, among other benefits, that the autoencoder offers.

I visualized the output side by side (Left: Direct Dimensionality Reduction. Right: Dimensionality Reduction using Autoencoder).

As you can see, some of the nuance in how close each sentence is in “empathy” to the other is lost in the direct embedding, but preserved with the autoencoder, allowing us to better understand how the LLM “embeds” this data and makes meaning out of it. For the image on the right, read each sentence from top to bottom, and check if you agree with the level of empathy that can be ascribed to each sentence (knowing of course that this can be subjective). Without performing a scientific analysis, I’d say that we can almost see a correlation between the embedding and the level of empathy (as indicated by the arrow on the right).

This is by no means, a comprehensive sample, and the nature of the output, its interpretation, and eventual utilization depend on various factors, including the chosen embedding model, but at least, this begins to give us some idea as to how an LLM may interpret inputs and outputs. The potential to build on this ability and its applications are what we are focused on at Relativ, my new startup, where we deploy customized AI models for personalized conversational analytics at scale.

​So, to recap, let’s do a quick summary of how autoencoders can be used to describe the inner workings of neural networks. Autoencoders can help by:

• Reducing data dimensionality to highlight essential features.
• Extracting and interpreting key features that drive network decisions.
• Analyzing reconstruction errors to understand network limitations.
• Providing insights into the transformations occurring within intermediate layers.
• Simplifying complex models to make them more interpretable.

Implications for Transparency, Understandability, Trust & Safety
One of the biggest drawbacks with the black box nature of LLMs is whether we can trust their outputs. This is particularly true if the Generative AI models are used in any capacity to provide advice, coaching, tutoring, or information that is necessary for learning and development. One such limitation of LLMs that is inherently present is termed as hallucination, which you can read more about in this article. While there are ways to mitigate this, the underlying question of why LLMs hallucinate, and how these large neural networks remain of interest. Unless we can “trust” the information provided by the AI model, which in turn translates to reliability and repeatability, and know that the information can be “safely” consumed, serious applications of AI to help us in daily life may be counterproductive.

​Another aspect of this is the question of “transparency” and “understandability” of AI models. This is a related aspect that governs the nature of the data used to train the neural network. The ways in which a neural network may learn information and fire activations may be largely dependent on the input data. Without transparent information on how the AI model is producing its outputs, and what data was used to train the model, it is difficult to truly “understand” what the output from the model means.

This is why researchers are so keen on understanding the inner workings of neural networks. Autoencoders are one such powerful tool available to us, and researchers are already starting to make progress towards this goal. At Relativ, we continue to focus on these aspects as the cornerstones of all our customized AI models. We know what data goes in, and we define what data comes out, in collaboration with our partners and clients. If you’d like to learn more about how we can help you integrate customized AI models to support your organization’s academic or business goals, please reach out. We’re always looking for people who care deeply about the positive impact of AI on our societies while keeping privacy, transparency, trust, and safety as our most important guiding rails. I hope you enjoyed reading this as much as I enjoyed writing it!

​In my first blog post about Generative AI and Large Language Models (LLMs), I explained their inner workings and why they are well-suited for adoption in both academic and business contexts. I’m building on this introductory blog by diving deeper into their inner workings in specific contexts. In this second blog, I will focus on the characteristics of LLMs that make them well-suited to help with career readiness preparation, from the grassroots stage (high schools, colleges), to later in your career, when you’re ready to land your dream role at your favorite company. We’ve already seen that Large language models (LLMs) are revolutionizing various fields, including interview coaching to find your first job, college admissions preparation, and preparing for job interviews while navigating various careers (experience a demo). Their unique properties enable personalized guidance based on individual performance, making them an invaluable tool in refining the human skills needed to be successful in any career. Let’s look at some of these properties in more detail and explore the science behind “why” LLMs are so well-suited to providing personalized guidance, and what constraints must be put in place to ensure a reliable output from these models.

Advanced Natural Language Understanding

LLMs possess advanced natural language understanding capabilities, allowing them to accurately interpret interview responses. They can comprehend nuances in language, identifying strengths and weaknesses in communication skills. The best reference for Large Language Models (LLMs) possessing advanced natural language understanding capabilities to accurately interpret interview responses and identify strengths and weaknesses in communication skills can be found in the paper "Improving Language Understanding by Generative Pre-Training" by Radford and Narasimhan. This paper introduces the popular decoder-style architecture used in LLMs, focusing on pretraining via next-word prediction, which enables these models to comprehend nuances in language and exhibit advanced natural language understanding capabilities. To take advantage of this capability, it is important to provide information and context to the A.I. model that are specific to your needs. There are several ways to perform this task, ranging from few-shot learning to fine-tuning, the scope of which is beyond the scope of this blog. Courses such as these from DeepLearning.ai can be extremely useful if you are new to this field.

Adaptive Feedback Mechanisms

These models utilize adaptive feedback mechanisms to tailor coaching based on individual needs. By analyzing interview responses, LLMs can provide targeted feedback, focusing on areas requiring improvement while reinforcing strengths. To better understand the science of LLMs and their adaptive feedback mechanisms to tailor coaching based on individual needs, you can refer to the following articles:

• AdaPlanner: Adaptive Planning from Feedback with Language Models: This paper introduces a closed-loop approach that allows LLM agents to adaptively refine their plans based on environmental feedback. It discusses in-plan and out-of-plan refinement strategies, a code-style LLM prompt structure, and a skill discovery mechanism. AdaPlanner has shown superior performance in various environments, particularly with few-shot learning.
• AdaRefiner: Refining Decisions of Language Models with Adaptive Feedback: This work presents a framework designed to enhance the synergy between LLMs and Reinforcement Learning (RL) feedback. AdaRefiner includes a lightweight Adapter Language Model that refines task comprehension based on RL agents feedback, without altering the generalization abilities of LLMs, while improving decision-making capabilities in and adaptable and efficient manner.

These two papers provide valuable insights into how LLMs can effectively utilize adaptive feedback mechanisms for tailored coaching based on individual needs.

Attention Mechanisms & Transfer Learning

LLMs leverage vast datasets to generate data-driven insights into interview performance. By comparing responses to successful interview patterns, they can offer actionable advice to enhance performance. One such dataset is the MIT Interview Dataset, which comprises 138 audio-visual recordings of mock interviews with internship-seeking students from the Massachusetts Institute of Technology (MIT). This dataset was used to predict hiring decisions and other interview-specific traits by extracting features related to non-verbal behavioral cues, linguistic skills, speaking rates, facial expressions, and head gestures. Modern day LLMs have some unique properties that allow them to exhibit similar capabilities with contextual information and new data, even if that data is not as comprehensive as the dataset described above. I briefly highlight two of these properties below:

• Attention Mechanisms: LLMs employ attention mechanisms, such as self-attention or multi-head attention, to weigh the importance of different words or phrases in interview responses. This allows them to focus on relevant information and provide more accurate feedback. In simpler terms, they consider all the words in a sentence as context and weight the importance of each word when processing another word, allowing them to discern meaning. You can read more about this in Google’s landmark paper here.
• Transfer Learning: LLMs often utilize transfer learning, where they are pre-trained on large-scale datasets for general language understanding tasks before fine-tuning on interview-specific data. This transfer of knowledge enables LLMs to leverage previously learned patterns and features, enhancing their ability to provide effective coaching by tapping into their general language understanding capability.
  

The result is that LLMs can be used to analyze conversational exchanges and responses of an end user, without the need for massive amounts of training data, if you can provide the model with sufficient contextual information on what to look for, in a manner that the LLM can comprehend (e.g. fine-tuning).

Real-Time Computation

One of the key strengths of LLMs is their ability to provide real-time analysis given structured data. This instantaneous feedback enables candidates to adjust their approach on the fly, improving their performance as they go. The architecture of LLMs, which are essentially complex neural networks, is optimized for efficient computation. These neural networks have already learnt a mapping between the input and output based on billions of parameters and are utilizing these learnt weights to perform mathematical computations at a rapid pace. This allows them to analyze conversational information such as interview responses in real-time, providing immediate feedback to users during these interactive sessions. Based on analysis and feedback, LLMs can help create personalized learning paths for career readiness preparation. They can even be tuned to recommend specific resources or exercises tailored to target areas for improvement, thereby maximizing their effectiveness.

How can Generative AI help my organization with career readiness?

In conclusion, the properties of large language models make them exceptionally well-suited for providing personalized career readiness preparation. Their advanced natural language understanding, adaptive learning mechanisms, data-driven insights and real-time computation capabilities offer invaluable support to individuals navigating the complexities of the pursuing their career goals. As LLMs continue to evolve, they hold the potential to revolutionize career development, empowering individuals to achieve their professional goals with confidence and competence. The reliability and consistency of the output from these LLMs is however heavily dependent on the quality of your input data and the precise definition of context that you can provide. At Relativ, we help organizations gather input data and create guidelines with sufficient fidelity for their A.I. models to infer “what good looks like”.  We help them experiment with their own data and understand how an LLM works with different contextual information, so they can expand these capabilities and begin to measure various skills that individuals exhibit during their conversational exchanges. When tailored to specific job descriptions, these customized A.I. models can give end users a competitive advantage by not only identifying the skills they require, but also helping them improve their performance on those skills through personalized feedback. Head over to relativ.ai or reach out to us to learn how we can help you deploy your own AI models, infused with psychology, and linguistics, to empower your organization and end users with the career readiness skills they require to meet the challenges of the future of work.

In my first blog post about this exciting area of research and application, I will aim to explain the inner workings of Large Language Models, and some of their characteristics, in a way that allows us to use a data-driven approach to decision making. The hope is that this information will allow the adoption of these AI models in our workplaces (and our personal lives), to augment our efforts, and help us be more productive, and efficient in the long run.

What are Large Language Models?
​
Almost everyone that has heard of Artificial intelligence, has also probably heard of the term “Neural Network”. They are a specific architecture which allows computers to learn the relationship between input samples and output samples, by mimicking how neurons in the brain signal each other. The first Neural Network, called a Perceptron, developed in 1957, had one layer of neurons with weights and thresholds that could be adjusted in between the inputs and the outputs. A fantastic introduction to this topic can be found here. Large Language Models (LLMs) are a type of Neural Network. In contrast to the Perceptron from 1957, some of these LLMs are infinitely more complex, with nearly a hundred layers and 175 billion or more neurons.

What do Large Language Models do?

One of the first practical applications of neural networks was to recognize binary patterns. Given a series of streaming bits (1s and 0s) as input, the network was designed to predict the next bit in the sequence (output).  Similarly, in very simple terms, Large Language Models can predict the next word given a sequence of words as input. If that is all that they can do, why are they so powerful and appear incredibly intelligent? The scope of this goes well beyond a blog post, but here is an incredible resource that will help you understand the inner workings of large language models.

Characteristics of Large Language Models

In the final sub section of this blog, I will attempt to highlight some properties or characteristics of Large Language Models. I will refer to these characteristics in various future blog posts, so the utility of these properties and their application areas can be better understood. It is these characteristics and properties that contribute to the ability of LLMs to appear to understand and generate human-like language.

• Word Vectors: Computers represent all information as bits. LLMs, in specific, represent words as numerical vectors. If you read through the link in the previous sub section on what LLMs do, you will already be aware that by storing these words as vectors, LLMs can capture semantic relationships (or meaning) between words. More importantly, LLMs operate in such a high dimensional space that they can compute the meaning of entire paragraphs and sections of text, comprising of thousands of words. This is what allows them to understand human requests and generate responses.
• Transformers: One of the key components of LLMs that can actively exchange information with humans is the transformer architecture. The transformer architecture allows parallel processing of information while also propagating information between sequential layers, allowing the model to learn complex relationships and patterns in the language (i.e. the relationship between words).
• Training Data: The top LLMs in use today have benefited from vast amounts of training data. It must be noted that these LLMs are trained on unlabeled text using self-supervised or semi-supervised learning methods. In the first stage, the model is trained on all the textual information that was available on the internet (sourced using crawlers or similar). With this data, the neural network learns to predict the next word in the sequence. In the next stage, the same model is given labeled data, of a much higher quality, with labels, but on a much smaller scale. The process is referred to as fine-tuning and use to generate contextually relevant content, in a format that matters to the end user, from the LLMs. Andrej Karpathy has an excellent podcast on how LLMs are trained, and offers valuable insights into how prediction and compression (e.g. zip files) can be shown to have a close mathematical relationship. It follows that larger the amount of data, and larger the model parameters, the better the ability of the model to predict the next word in the sequence. Using this base model to perform fine-tuning inherently guarantees better results (subject, of course, to the dataset and labels used in the fine-tuning phase).
• Fine-Tuning and Prompt Engineering: As I previously mentioned, fine-tuning is essential for LLMs to perform specific tasks in a consistent and effective manner. Fine-tuning helps optimize the model's performance for a particular task such as analyzing sentiment or generating summaries in a specific format. When used in combination with other techniques such as prompt-engineering, the ability of a fine-tuned model to adapt to a new task and context can be quite magical and have a huge utilitarian value. A simple online search will yield several great resources on prompt-engineering, which is a highly sought after skill / emerging field. I strongly recommend that anyone interested in using GenAI for your personal or business use, to experiment with these techniques and solve relevant challenges that you encounter on a daily basis.
• The Black Box Problem and Emergent Behavior: So far, we’ve focused on the things that we truly understand about LLMs. But it is also important to recognize that the sheer size and complexity of LLMs pose a significant challenge to our understanding of their true inner workings. The “black box” problem draws attention to the fact that the logic behind the decision-making process of these large neural networks is not easily traceable, and hence, not well-understood. This article in Nature highlights the black box problem and offers a good balance in perspectives on why this problem matters, and how we can continue to work around it while receiving the benefits that AI offers. On a related note, LLMs exhibit something known as “Emergent Behavior” i.e. as the size of the models gets bigger, they exhibit capabilities that they were not trained to perform. It is this characteristic that makes experimenting with LLMs to solve business challenges a particularly exciting area of interest.

What can Large Language Models help with?

At Relativ, we have been experimenting with Deep Learning, Psychology, Linguistics, and Large Language Models, by qualitatively assessing the generated text across thousands of interactions, and continuously instruction-tuning these models to quantize their output. We have learned that when LLMs are combined with proprietary algorithms and curated data, their output can be transformative and insightful. When used in thoughtful conjunction around the end-user experience, as well as the intended business or academic outcome, we are starting to see some very promising results. We are already beta-testing these models in recruiting, sales, learning and development, retrospectives, and career readiness where early adopters are reaping the rewards of experimenting early, and gaining a competitive advantage from the learning that occurs.
In the next series of blogs, I will attempt to describe how Relativ's models are being used in each of the above fields, and why they can be a game changer in the long run. I will refer to the characteristics of LLMs highlighted in this blog entry, for continuity and chain-of-thought (pun-intended), throughout this series. In the meantime, head over to relativ.ai or reach out to us to learn how we can help you deploy your own AI models, infused with psychology, and linguistics, to help you drive business outcomes.

It's been a while since I had the time to update this blog. I was recently interviewed by Superbcrew, an online business and  technology review magazine.  I took the opportunity to elaborate on the effectiveness of Mursion and what the future holds in store for us. You can read the full interview by clicking on the image below.

There are several techniques that are used for soft skills training in industries such as customer service, corporate leadership, healthcare and sales among others. The primary goal of any such training is to enable professionals to deal with difficult situations they may encounter in the real world. The theory behind training is that exposure to such circumstances will help enable quicker and better decision making. For several decades, live role playing has been seen as an effective methodology for soft skills training. There are several advantages that role-playing offers in contrast to other passive training methods.
 
•   It provides a safe space in which learners can make mistakes and try out new strategies of problem solving without being subject to the risks that are associated with such situations in the real world.
•   The active nature of the learning, requiring 2 or more-people to be involved automatically lends itself to new learning stimulus that includes cultural diversity, language, non-verbal reactions, personalities and other human characteristics that are not independent of the learning goals.
 
For role play to be effective, there are certain pre-requisites that include actual physical locations, authentic real-world scenarios, and good actors / consultants all of which reinforce the concepts of situational plausibility and place illusion - two key constructs whose origins and importance I have described in a previous blog. In addition to these pre-requisites, the role playing scenario must be recorded so that participants can watch and reflect upon their performance. It is this combination of being in a realistic stressful situation and reflecting on the performance after the role-play that makes it effective - in other words, experiential learning occurs when role-play based learning is executed well. However, all the pre-requisites of role-playing make it logistically expensive and difficult to execute in a manner well-suited to learning. It is here that we can harness the power of virtual reality and avatars, to scale and amplify an existing training methodology. This is the essence of what we do at Mursion. In fact, several years of studying the VR alternative to role-play using avatars has led us to believe that this methodology is not only easier and cost-effective to scale while increasing outreach, but also maybe better and more effective than traditional role-playing techniques. 

Firstly, for learning to occur, it is important for one to experience a certain amount of discomfort. With traditional role-play, it may be difficult or awkward for an inexperienced role-player to push the learner outside of his/her comfort zone, without breaking the illusion that the situation is real. It takes this sort of pushing to trigger the mistakes that are so costly in real life. If a simulation is to inoculate the learner against the emotional reactions that trigger bad decisions, this pushing is essential. In VR, the mask of being behind an avatar enables the role-player to push the learner to take risks. The role-player is never visible to the learner – the learner never catches the role-player’s gaze; they never “connect” as humans. In fact, studies have shown that social influence is greater during interactions where a human is behind the avatars during these conversations. Thus, the role-player feels liberated to push the learner in ways they would find very hard to do repeatedly in a live context.

Secondly, in the traditional sense, a role-player may not physically resemble the individual(s) who pose the real-life interpersonal challenges. The concept of standardized patients in healthcare are a prefect example, where it is very difficult to recreate a pediatric scenario of a 5-year old child with a medical condition for obvious reasons. With VR, avatars can be built to look and sound precisely like the people that you encounter at work, and with whom you may have negative interactions. This becomes critically important when issues of age, race and gender matter to the simulation. Additionally, environments can be customized so that it feels as though you are having the conversation in the same work environment where the conversation takes place. All of this allows us to meet the pre-requisites needed for effective role-play in a cost effective and logistically convenient manner.

At Mursion, we focus on blending live human performance with AI-driven avatars to blur the learner's sense of what is real and what is not. Machine learning is used to gather role-player intent, digital signal processing (DSP) technologies are used to mask the role-player’s identity and advanced low-cost networking paradigms support bi-directional communication between multiple learners and multiple avatars within a single simulation session allowing us to scale training to the maximum. We are often asked the question of how this approach is scalable if a role-player is always needed in the system. The answer is multi-fold:

• A single role-player can play several avatars simultaneously on screen. This allows us to create simulations such as pediatric scenarios where a learner (doctor / nurse) must deal with a complex situation involving parents and a child. The possibilities of such scenarios across other verticals such as leadership training are many as you can imagine.
• Multiple avatars can be tailored to match any age, gender, or ethnographic profile on screen, allowing us to create scenarios around diversity/inclusion and implicit bias and study them in context.
• Multiple learners can remotely login to the same simulation session with multiple avatars and take turns practicing various strategies while others shadow the learning. And in each case, the very same simulation can play out very differently exposing learners to the concept of multi-exemplar training - another concept that I discussed at length in a previous blog.
• The role-playing sessions can all be run remotely. Role-players are simply subject matter experts who can “inhabit” the avatars from the comforts of their home. Simulations can be designed so the role-players can simply be themselves to pose the learning challenge.  Learners can experience the simulations from laptops, using VR headsets or on their mobile devices wherever they are located. In essence, this combination of VR and AI has helped turn role-playing into an anywhere-anytime activity.
• If we were to create purely AI-driven avatars, we would require a very large amount of data to create authentic interpersonal exchanges. In addition, we would need continuous programmatic content creation rather than having a flexible platform where content can be tailored in real-time. In our particular case, it is better to place the human at the end of the chain rather than in the middle (see relevant blog).
• And most importantly, all the data from the simulations is recorded for review and reflection eliminating the need for extensive logistics and planning. We are currently building algorithms for behavioral analytics in the context of these complex simulations and have identified new techniques that allows us to tie them back to ROI measures for organizations.   More on this to follow.

​In the last 4 years at Mursion, we have seen evidence of learners trying their hardest to beat the simulation. In their effort to win, learners take conversational risks that they would never take in a live role-play. These strategies sometimes work, but more often than not, they fail. And it is this failure that prompts learners to reflect, learn and be better prepared. If you are trying to incorporate cutting-edge VR & AI simulations into the professional learning and development curriculum at your organization, come visit us at Mursion. And with that, I’d like to wish all of you a wonderful and prosperous 2019!

In the last few years, Virtual Reality (VR) and Artificial Intelligence (AI) have been two topics of huge interest across various industries. Both these topics are so broad in terms of their application areas that it is almost difficult to find an industry that would not benefit from the use of one or both of these technologies. Application areas range from entertainment, to analysis, to training, assessment, and everything else in between. But what if we could combine the power of these two technologies, augment it with human reasoning, and create applications that lend themselves to entertainment, analysis, training and assessment among other things? We open a new door that now requires insight into understanding human psychology and how people behave when interacting with these technologies. More importantly, it becomes critical to design how these technologies manifest themselves when interacting with people.

​I’ll try to explain this concept within the boundaries of what we currently do at Mursion - a VR and AI based training platform designed to help acquire skills through experiential learning. Mursion was built on the foundational principle that difficult conversations in real-life can be simulated before they occur, to safely inoculate against unpredictable behavioral changes in the presence of stress. Many of us would agree that we have been at the receiving end of such events more than once in our lives - be it at work, at home, or in a new geographical location where socio-cultural practices differ from our individual perceptions of the norm. While experiencing Mursion, learners’ beliefs that the simulation is real and is happening in their physical space are continuously reinforced - this concept was well studied and published by Mel Slater (described further in one of my previous blogs). This is achieved by carefully stimulating the visual and auditory senses of the learner using Virtual Reality and Artificial Intelligence. 

​To do this, we follow certain key principles that we have learned by analyzing data (no personally identifiable data is ever collected) or seeking feedback from hundreds of learners experiencing the platform:
 
(i) We design simulation scenarios as short vignettes, no longer than 7 - 10 minutes long. The content of the scenarios is both relevant and of high impact with respect to the learner. Virtual Reality allows us to position the learner in this situation by helping modify the visuals to match the context in which the vignette is occurring. For example, a conflict with a co-worker must occur in the workplace of the learner and not in a generic setting. We go to great lengths to customize the simulation content and model these using reference photographs that are relevant to the vignette being designed. 
(ii) We use a combination of Human Reasoning and Artificial Intelligence to create avatars (digital representations of humans) that blend seamlessly into the simulation setting. While the mechanics of this blending are proprietary, spoken dialogue on the Mursion platform is not driven by AI (I’ve highlighted the reasons in a previous blog). In fact, we have learnt from data that language, accents, and even vocabulary during these highly engaging simulations play a huge role in reinforcing the authenticity of the interaction. The more authentic the simulation, the better are the learning outcomes.
(iii) All scenarios have discretely measurable objectives / goals for the learner. We leverage this to implement a 360 degree feedback mechanism and rely upon reflection as a powerful learning mechanism. Learners receive a video of their performance in simulation that can be shared with peers, coaches and others to seek feedback while also watching themselves and taking notes that night help them get better in subsequent sessions. We are now testing a system that uses Artificial Intelligence to analyze the audio and video streams of the learners to present the learners with highlights of their interaction - micro interactions in which they had a positive impact or a negative impact on their avatar counterpart during the simulation. While we are still studying and refining our approach to this analysis, there is sufficiently positive evidence that this data can be used to provide workforce analytics that may help re-organize teams for maximum productivity/efficiency and so on.
(iv)  We design simulations in a manner that does not reward rule-governed behavior. Once again, experience and data has shown that there are multiple ways to achieve a certain outcome during difficult human conversations. My good friends, advisors, and mentors William Follette and Scott Compton have been at the forefront of helping us study the science behind this concept. As they would put it - “We can only tell you certain things you should not do during a conversation (e.g. do not dig your nose!) - everything else that you try in order to have a positive impact on your counterpart is fair game”. What is more, every conversation and every person is different - it is critical that you assess your own impact on others and change your behavior (verbal and non-verbal) accordingly until you achieve the desired outcome.
(v)   And finally, we rely on the concept of multi-exemplar training - learners are exposed to several simulations around particularly difficult situations, but each simulation may have a different learning outcome. No two simulations are ever the same (unless it is required for an assessment). Learners see a variety of demographics in the form of avatars, with each one posing a unique learning challenge. The more simulations a learner performs, the more prepared they can be to handle stressful conversations that may occur in the real-world.
 
While there may be several other ways that AI & VR can be be combined depending on the context of the application, we, at Mursion, have encountered the need to study human psychology when using these powerful technologies. We have also learned that blending human reasoning with VR and AI can help create a technology that can be used in a plethora of ways. I will try to address some of these in a future blog. But without the science that helps us understand behavioral psychology when interacting with technologies, it is difficult to truly harness the impact they can provide. If you would like to learn more and see if such technologies may be relevant or impactful, please drop us a note and come see us at Mursion. We’re in the bay area and always happy to entertain visitors! 

Those of us creating engaging and effective solutions in the Learning and Development (L&D) space, have invariably heard the acronyms VR (Virtual Reality), AR (Augmented Reality), MR (Mixed Reality) and most recently, XR. While the first three acronyms have specific meaning in a traditional sense, the ‘X’ in XR is interpreted differently by experts in the field. One such interpretation is “Cross Reality” spanning the use of Virtual, Augmented or Mixed reality - Wikipedia has a detailed description on the history and origins of this term.
 
So which one of these “pseudo” realities is the most effective for learning and development? The answer to this is dependent on several factors, the foremost of which lies in the question itself - determining what “effective” means in the context of the specific learning and development. For a methodology to be effective, it must first be engaging. In other words, a learner must feel the desire to be accepting of the proposed methodology as a learning tool. Next, the methodology should create a retention of knowledge. And finally, the methodology must instill confidence in a learner to apply the knowledge in real-world circumstances. With these three goals in mind, we can explore the use of VR/AR/MR or XR (a combination of realities) in achieving the desired learning and development goals.
 
In my first blog, I highlighted some basic design principles that must be followed while creating VR experiences - principles that rely on the effective replacement of the senses in order to create engaging applications.  AR and MR differ in the fact that the former is used to overlay / augment the real world with additional information and the latter performs a similar function, with the additional burden of anchoring the information to a physical object in the real world. There are plenty of mechanisms (and applications) in which any of these realities can be achieved using a combination of multiple sensors and visual displays that are beyond the scope of this blog. What is of importance is determining whether the application in question benefits the learner in achieving the three goals established above. To illustrate, it may be best to visualize what these different technologies look like in the context of learning and development - soft skills are one such area of critical importance (highlighted in a previous blog), independent of the industry in which an individual may be employed.

Computer generated props can be displayed (rendered) on a computer screen, in a VR device (headsets such as the Oculus Rift or HTC Vive), or through an AR display (devices such as the Hololens). In the first image on the left, everything that is displayed is virtual - the classroom, the trees, the books and the avatars. This simulation is great in order to give the sense that a learner is in a classroom and is most effective when rendered via a VR device. Depending on the level of immersion desired (different levels of immersion create varying degrees of engagement in different learners), one may choose to simply use a large television screen or computer screen to experience this simulation. The second image shows a “hologram” of the girl avatar rendered via an AR display (in this case, the Hololens). The chair in this image is real while the girl avatar is computer generated. More importantly, the girl is anchored to the chair. The “anchoring” construct involves complexities not described here, but for simplicity of understanding, we may consider this “Mixed Reality”. If well-implemented, a swiveling chair would result in the girl avatar spinning as if really seated on it. One can imagine the use of this modality to create a virtual learning companion in the natural learning environment (e.g. in homes of students). The third image shows the girl anchored to the chair, but also an orange-red circle. This circle is a cursor that allows one to interact in the “Mixed Reality” environment. However, the cursor is not anchored to any object in the real world - it is simply an overlay (via the display) on the real world. This may be considered a prop in “Augmented Reality” and could change its color, or description as it hovers over different objects in the “Mixed Reality” setting. Such informational overlays could be used to assist a learner in achieving their goals and gradually phased out to increase the learning challenge. As you can see, the terms AR, VR and MR are all unique, yet intertwined.
 
At Mursion, we focus on using simulations to foster empathy and induce positive change in human behavior. While we await the adoption of these XR technologies at scale, we keep our platform compatible with existing delivery modalities - television screens, laptops and mobile platforms - while continually striving to create the most effective and engaging content for the delivery medium. We prefer to think of XR as “Experiential Reality” - a simulated reality that facilitates learning via experience, independent of the medium in which the simulations are delivered. If you’re wondering how best to incorporate simulations into learning and development, get in touch with us or better yet, schedule a visit to beautiful San Francisco and come see us at Mursion!

Alexa, Siri, and Google Assistant are household names today. Advances in Artificial Intelligence (AI), more specifically machine learning and natural language processing, have empowered us to use voice to interact with machines. This adds a new dimension to existing literal and gestural interfaces and potentially overcomes the complexity of user interface design when navigating complex / deep menus. The simplest example is that of using Alexa to play a song - a user interface on iTunes or similar would require that you type the name of the song in the search field, apply any filters to narrow the search, and finally selectively push play based on the results. Instead, these actions are replaced by your voice, eliminating the need for a sequence of user inputs thereby speeding up the interaction. But perhaps the most important aspect of voice-based interactions is that it has eliminated the need for proximity with the machine and / or it’s physical user interface. From a psychological perspective, it plays into human nature - we primarily communicate with other humans through voice, body language, and gestures. Why can’t we do the same with machines ? Note that the word “communicate” is very different from the word “interact” - and it is here that we begin to encounter barriers with all aspects of AI related to Natural Language Processing (NLP).
 
To help better understand the difference between interaction and communication, let’s use the following example. Try asking your favorite voice assistant to order a coffee:
 
"Siri, can you place an order for a cappuccino using the Starbucks app? "
 
The results will likely be locations of the Starbucks stores that are in your vicinity. More specifically, the result would have been appropriate if your question was "Find the nearest Starbucks " In order for you to achieve your intended goal (ordering a cappuccino), you must continue issuing voice commands or interacting with your phone until the above task is complete, but the whole process feels transactional in nature. They are short, simple constructs, with each subsequent instruction having a definitive outcome.
 
But what if you had this same exchange with a friend or co-worker? You will likely begin with a discussion of what the outcome should be, exchange complex constructs during your interaction and never worry about the order in which the exchange of information occurred. In addition, you will have no problem dealing with variations in accent, language, grammar, or even if the task was completely inter-twined with another. 
 
It is obvious from this very simple example that you can interact with Siri or other voice assistants but can never really communicate with them (at least not today). So what causes this barrier and can it be overcome?
 
Communication is a very complex construct. When we communicate as humans, we perform several conscious and subconscious tasks such as reading the other person’s facials, gestures, body languages and tone of voice in the context of the interaction so we can best reciprocate to achieve a desired outcome. Machines have no such knowledge and in particular, are devoid of contextual information. Several researchers are investigating the possibility of giving machines the ability to perform some of these tasks, but the challenge lies in the nature of the machine to acquire this ability. Over the years, this has been termed as “Artificial Intelligence” and encompasses several sciences of which machine learning has gained popularity since it provides machines with the capability to generalize taught constructs. There are however two things to bare in mind:
 
•   The word “Artificial” refers to the fact that machines can be made to “appear” intelligent and are not truly intelligent.
•     Machines needs a lot of clean data in order to “learn”.

To illustrate, if a relationship A <> B and a relationship B <> C are taught, humans are capable of deducing the relationship A <> C. Not only that, humans are also capable of deducing the reverse relationships B <> A, C <> B and C <> A (see example to the left). In fact, it is this ability of ours to learn via extrapolation / deduction that makes us the dominant species on the planet. For a machine to learn all these relationships, humans must provide data that includes both positive and negative examples of all of these relationships. In other words, humans must provide a machine with sufficient data for it to learn to “compute” all these relationships. And this is the AI effect - if a machine can compute a relationship, it is not truly intelligent. The advantage of machine learning is best seen when a problem is complex and non-linear and hidden relationships between the input and output need to be discovered. And this is where the biggest barrier in teaching a machine how to communicate exists. The subsets of inputs and outputs are infinite.

​Take for example, the common situation where employees engage in a 5-minute conversation with a manager about a workplace dispute. My previous blog post addresses why soft skills are so critical to increasing productivity and efficiency in the workplace and how training can help better handle such situations. The workplace dispute could be around any topic involving budgets, processes, methodology or even personal in nature. One can only imagine the number of unique ways in which this conversation may play out in real life depending on a number of factors including but not limited to personality types, context surrounding the conversation, history and relationship between the people, etc., re-iterating that infinite space of inputs and outputs during communication.
 
Let us for the sake of argument, assume that AI can be programmed to hold an extremely engaging conversation within constrained boundaries for 5 minutes. This readily lends itself to a training application where learners can practice their soft skills. Even in this case, the machine will need sufficient data from real conversations in these areas to communicate effectively with a learner. More importantly, the machine will require contextual information and read the learner’s body language, tone of voice, and other non-verbals to effectively engage them in the training. How is it possible to provide the machine with this learning data unless we have a database of such conversations readily available to us ? Every time there is a new pathway of conversation (data that the machine has not seen and hence cannot deduce a relationship), a human will need to provide the machine with both positive and negative examples to incorporate this pathway into its repertoire. In other words, there is always a human in the loop. The question we ask ourselves is this - where do we want to place the human in the loop ?
 
At Mursion, we combine the computational power of AI with that of human reasoning to create seamless, engaging and customizable simulations that allow learners to practice soft skills. By placing the human at the end of the chain, we avoid having to teach the machine the complex art of communication - instead, we focus on harnessing the power of artificial intelligence to drive digital representations of the human (avatars) thereby reducing the cognitive load of the humans that inhabit the avatars. The end result is a simulation platform where subject matter can be rapidly customized without the need for re-programming. When combined in this manner, virtual environments and avatars reinstate the constructs of situational plausibility and place illusion (Slater, 2009), leading to the suspension of disbelief in learners. This creates a very realistic, effective, and safe training environment. To learn more about how this training can be applied to organizations and how results translate to real-world performance, come visit us at Mursion - and as an added benefit, you can walk to the Golden Gate bridge after your visit!

Every year, billions of dollars are spent on professional development and training across various industries.  Training ranges from the development and mastery of hard skills and subject matter expertise to soft skills that support better productivity and create stress-free environments in the workplace. While technical skills across industries are easily assessable, the impact and measurement of soft skills is a relatively hard topic to address. A recent surge in awareness around this area has resulted in research that shows the staggering impacts of soft skills training. Skills such as time-management, problem solving, teamwork, conflict resolution, and communication have a direct impact on a company’s bottom-line by improving teamwork, increasing employee engagement and reducing absenteeism to name a few. More importantly, soft skills are transferable across industries, making the return on investment (ROI) on such training very significant for both the company and the individual learners who invest in it.
 
Let’s begin with a popular example of a soft skill - active listening. Active listening involves paying close attention to a person, people, or event with a view to hearing, interpreting, and understanding the entirety of the message being communicated. Human parity in speech recognition is estimated to be roughly 95% (i.e. a native speakers understand 95 out of every 100 words they hear in that language). I plan to address more about this statistic and it’s impact in relation to Artificial Intelligence and Natural Language Processing  in future blogs. While the human parity number is remarkable, research suggests that we remember only about 25 - 50 percent of what we hear.  What if we were able to increase this percentage ? For a passive task such as watching a movie the additional information gleaned may allow us to better understand the underlying plot or director’s motives. For an active task such as one that involves a discussion among colleagues or teams, we would have more information to process, allowing us craft a better reciprocal response. By increasing the amount of information absorbed and disseminated thoughtfully, we create a feedback cycle that multiplies productivity.

Often, professional development or adult learning around soft skills tends to involve components that are rule-governed. A quick Google search will reveal that behavioral traits such as nodding or shaking your head in agreement, making eye contact, smiling etc. are all recommended to help establish rapport and engage with counterparts during conversation. There is a fundamental problem with this approach. We developed body language and gestures as a means of communication far before the power of speech evolved in us. On one of my visits to meet with Sandy Pentland at MIT (an advisor to us at Mursion), we began discussing how computers through machine learning / deep learning were capable of analyzing human conversation and predicting outcomes. If we think about the purpose of gestures and body language on an evolutionary timeline, we begin to see that they are an attempt by humans at turn-taking. They acknowledge to the other person(s) that the information presented is being absorbed, and specific gestures convey that we are ready to contribute to the conversation. Put in the context of active listening, it follows that these signals are subconscious in nature and are causal i.e. the body language or gestures occur because we are actively listening and not vice versa.
 
Sandy, in his book Honest Signals, presents very powerful scientific and data-driven evidence of how non-verbals can be analyzed to very successfully predict outcomes in human conversation. He goes on to explain how understanding and measuring these “signals” can help create a very successful and productive workforce. Understanding such signals however requires training of the conscious and subconscious mind. And it is only with continuous practice that we begin to build the relevant skills that are necessary for this task. Measuring performance while building soft skills and receiving feedback in the moment are critical aspects that allow positive behavioral change. Both of these are difficult to do in real life, since situations do not lend themselves to scientific measurement and providing constructive/tough feedback in the middle of a conversation (another soft skill !) is not easy to do without practice. It is here that we can harness the power of virtual reality and simulations - the premise of our work at Mursion. If you’re interested in soft skills development, pick up a copy of Honest Signals and come see us at Mursion to see how we create authentic and powerful Virtual Reality simulations for various soft skills and measure human performance to help professionals master their craft.