How will generative AI transform the Expert Network space?

A topic that fills me simultaneously with huge excitement at the endless possibilities, and trepidation at the pace of acceleration. On the one hand “its like having an army at our disposal to rapidly convert content into usable data” to quote Ross, from our data team. On the other, the power of the new tools and the multitude of applications can be overwhelming.

What will it boil down to for our sector? Well, I asked ChatGPT to provide a TL:DR on the topic

This sounds promising, but I want to share my own perspective on this topic as a founder of Techspert, an AI-native Healthcare-focused expert network. After all, human opinions still matter! (But, I would say that). Here are some of my musings on how AI will influence the future of expert networks.

Are Large Language Models (LLMs) Relevant for Expert Networks?

Yes, absolutely. Techspert is not unique in that we're sitting on a mountain of unstructured language data, in transcripts, emails, surveys, and bios, to name a few. “There is gold in them hills” and language models like GPT-4 are new picks, shovels and turbo-charged mining equipment that can extract it. On the face of it LLMs are a great match with the needs of expert networks to improve the match of insights to customer needs. What’s more, humans already play a key role in vetting in the Expert Network model, which can combine with LLMs to achieve 100% accuracy, even when LLMs themselves fall short of 100%. There is the potential for revolutionary changes!

Will AI experts replace human experts?

No, but they will complement each other. Our customers come to us because they want perspectives and opinions based on human insight and experience backed up by their professional qualifications and roles. Much of the insights gained are simply not available online - they are only stored in the grey matter of the expert. This is the entire premise of our industry and the reason we give an edge to our customers in their decision-making.

Similarly, a consensus view from AI, that aggregates across all online views doesn’t mean it's the most useful view. Gaining insights and views across a diverse range of opinions to understand where and why they differ is important for decision-making. The average view about a product is not as useful as understanding the outlier views and why some people really loved or hated it - it tells you more about where you could take that product for more success. Needless to say, an AI-generated review of a bed would also not help anyone for the same reason!

On the other hand, LLMs have now proven they can pass many medical examinations and AI therapists have proven useful functionality in providing meaningful counseling applications. In theory, chatbots can absorb human insights and be trained to respond in the same way that a 'Surgical Oncologist' would do in a given conversation.

So it depends. If you are interested in asking questions such as "Who are the leading companies in this field" then an AI expert may give similar responses to human experts. But when deeper insight is needed, AI falls short. For insights from an on-the-ground perspective of how tools and treatments are used, human insights will always be essential. AI and human insights can therefore be a complementary combination, in the same way that secondary research is already complementary to primary research.

Will generative AI reshape how clients consume expert insights?

Yes, but the transformation will likely occur through a series of incremental changes. Currently, clients engage with expert networks through email requests, platform bookings, video calls, transcripts, and survey responses etc. This process is riddled with pain points and is very time-consuming to get to the point of synthesizing the gathered information into meaningful conclusions.

In a hypothetical scenario, interactions could become purely conversational with an AI chat interface, where the AI summarizes and delivers final results to clients. Taking it a step further, adding in visual AI you could even see fully-fledged human-like assistants handling customer requests. However, this theoretical (slightly dystopian) vision is still a distant reality and might not even be the most efficient or convenient approach. For instance, consumer applications like ordering takeout or booking a holiday are not conversational interfaces, and it's unclear that managing knowledge and insights across an organization would benefit from such a method.

Nonetheless, incremental changes are happening already. One immediate application of large language models is the instant generation, translation, and summarization of accurate transcripts. This low-hanging fruit will likely be addressed swiftly, along with many other bite-sized improvements within the interface.

Additionally, advancements in search technology could make 'self-serve' models of expert identification more feasible and effective. These have not proven accurate and convenient enough to date and there is a graveyard of attempts. But provided the expert network has access to a sufficiently large pool of experts, the application of the latest search technology could solve the convenience and accuracy pain points. (Techspert covers over 200 million experts and offers an ‘Omnisearch’ feature, the development of which is being rapidly accelerated by AI tools).

Will generative AI transform the operations of Expert Networks?

Undoubtedly, yes, and fast! The benefits of generative AI are too significant to ignore. For example, entity recognition can convert unstructured biographical data and transcripts into structured fields that can be accurately matched and searched – a process that is already underway at Techspert. As these tools become more accessible, their adoption will increase. Converting mountains of content into gold, that improves match quality, speed and volume of results.

However, accuracy remains a crucial consideration. Expert Networks are accountable for the validity of the experts they present, ensuring they possess the claimed experience and qualifications. Maintaining the highest level of service and standards requires human vetting. With reports of fraudulent and fabricated expert panel responses to surveys reaching as high as 70%, maintaining quality is increasingly vital and means human oversight will remain critical.

Who will benefit?

The introduction of new tools and their capabilities will revolutionize the quality, speed, and convenience of expert network services. Delivering better experiences for customers and experts alike. This transformation will benefit users, and likely expand the market into broader segments of market research and the wider knowledge economy. With firms that leverage large datasets with the latest tools, well placed to benefit.

At Techspert, since we focus on the healthcare sector, with disparate publicly available information, compared to LinkedIn-dominated verticals, we can drill down further to understand expert’s specialties, roles, and experiences in powerful new ways to deliver a higher level of specificity than previously possible, while continuing to deliver experts they can’t be found anywhere else.

Why could progress be slower than expected?

Experience suggests that the rapid ascent up the hype curve is often followed by a trough of disillusionment before reaching a plateau of incremental value gains. Particularly in the B2B realm, each organization has unique and complex needs for knowledge acquisition and processing.

One-size-fits-all solutions tend not to work, and it will take time to integrate and roll out incremental improvements that save customers time without compromising quality or compliance. But the hype isn’t overblown, big changes will happen in our industry just as in most others!

The space is also moving much faster than regulatory frameworks can handle, and this will likely be another important bottleneck to adoption and development.

How does compliance factor in?

Compliance is crucial. Human insights wield tremendous power, but ‘with great power, comes great responsibility’ is certainly true in our sector. To be specific, certain information – such as confidential, proprietary, and non-public information related to public companies – should not be shared and regulations concerning data privacy, anti-bribery and many others, must be adhered to. Generative AI adds another complicating factor into the mix, with regulatory frameworks around AI moving very quickly.

However, this doesn't mean working with LLMs should be halted. Engaging with OpenAI’s API is contractually similar to sharing data with other third parties, but it is, of course, critical to ensure that any sent information isn’t used to train their models or resurfaced. OpenAI’s API terms and conditions update on 1st March was very welcome in this respect. It is a rapidly evolving space and firms will need to adapt rapidly to keep up with the pace of change. Generative AI could even assist with compliance by detecting fraud or suspicious activity, for example.

What are the implications for Techspert?

As an AI-native firm, Techspert has used language models to predict expertise since its inception, with a team of leading computer scientists in Cambridge. Our unique approach to finding experts – using search technology covering millions of experts rather than relying on either internal curated panels or LinkedIn – positions us perfectly to harness the latest tools for accelerating our capabilities. We eagerly anticipate unveiling exciting product announcements this year and embarking on a transformative journey for our sector. Stay tuned...

If you’re interested in learning more about how we’re using the latest generative AI to deliver more value to our customers and experts, or would like to experience it for yourself then get in touch or submit a project here.

Or if you’d like to chat more about how Generative AI will affect this industry and discuss where I’ve undoubtedly got it wrong, then drop me a message.