Artificial Intelligence seems to be taking over the world. It’s everywhere we go, and everyone is talking about it. Yet for all its promise and incredible potential, something is missing. Artificial Intelligence needs your Human Intelligence. Your, as in you. The reader of this blog. You as a representative of your business, regardless of your technical knowledge. What do I mean by this?
Artificial Intelligence seems to be taking over the world. It’s everywhere we go, and everyone is talking about it. Yet for all its promise and incredible potential, something is missing.
Artificial Intelligence needs your Human Intelligence. Your, as in you. The reader of this blog. You as a representative of your business, regardless of your technical knowledge.
What do I mean by this?
Before I get there, I’d like to show you three cutting-edge AI generated videos which Meta, the company that owns Facebook, published a month ago.
The first one is of this red-faced monkey playing with a sailboat in a hot spring. Pay close attention to the monkey’s fur and the reflections. Where the fur is wet and where it is dry. Only a few months ago, AI was struggling with generating realistic hands and fingers, and now it's capable of this.
Generated by and copyright of Meta
The next one demonstrates incorporating a real photo, shown in the top right, into a generated AI video. Looking at, for example, her cheekbones, the AI did a really good job, and that’s also not bad for a cheetah. It’s obvious how this tech can be used in future. Build your own story, with you acting in your own blockbuster.
Generated by and copyright of Meta
The last one shows a real video, of a poodle chewing on a stick, being augmented by AI. She’s given a onesie, transported to a royal garden, and turned blue.
Generated by and copyright of Meta
I wonder what your reactions to these videos are. They’re undoubtfully impressive. Are you amazed? Inspired? Enthusiastic? Did these videos get you on the edge of your seat?
ChatGPT was released in November of 2022, nearly two years ago, and since then, we’ve seen countless impressive demoes of what Generative AI can do. While those demos have become more impressive, I wonder if some of your reactions aren’t more negative?
Overwhelmed, exhausted, disillusioned? Perhaps even annoyed or jaded?
The reality for most of us is that there is a big disconnect between the promise of AI and the actual impact it has on our day-to-day lives. While Silicon Valley is talking about Artificial General Intelligence and the entirety of society changing, I still struggle to get Siri to properly understand me when I ask her things.
This is even worse within the business context. AI is talked about everywhere these days. From personal experience at clients, it’s hard to get budget for a project if it doesn’t have some element of AI attached to it. Yet often these AI initiatives don’t seem to have a clear goal. AI is ‘magic’, so we can add it to any system and get a ‘magical’ solution.
To a large degree, AI – which for this blog’s purpose refers to Generative AI – has become the hammer in the hammer and nail metaphor. It is so powerful and so easy to apply to any domain, that it is forced into every problem. What is the result of that? Generic chatbot after generic chatbot. And while it is impressive that I can ask a chatbot to trigger some action for me, and be understood, I was also fine with just pushing the button myself.
But it doesn’t have to be like this; and by assuming Generative AI provides a generic solution to all problems, we’re reducing its effectiveness and value.
I’ll now share a different AI showcase. One that isn’t nearly as impressive as the monkey video, but which I think is rather smart.
Google released an experimental AI platform, called Google Illuminate, which takes research papers and uses AI to summarise their content, and then present it in the form of a panel discussion. So, it does three main things:
Let’s take a look.
There is a reason panel discussion is a popular way of relaying information, and Google now allows that to be applied to any research article. AI has already shown that it’s good at summarising content, but the format and voice adds another level of value to it.
Your turn
Of course, Google could create this. They’re Google! They’re one of only a handful of companies with their own foundational LLM model, in the form of Gemini.
What if you had this idea? What if you felt that sharing research content with students or researchers could be more effective if it was presented as a panel discussion? What if you wondered whether AI could help with this? How would you even go about finding out if that’s feasible?
Well, how about you give it a try.
This video uses OpenAI’s playground to try and replicate what Google Illuminate does.
One of the most amazing things about AI is that it has significantly decreased the effort and knowledge required to test ideas and create Proof of Concepts. Ideas can be validated by the very same person who has the idea.
We created a digital clinical case management tool for medical doctors. The one doctor told me that he has all these notes on patients, and he was wondering what AI could do with that: what type of help or suggestions could it provide? He asked me, as an AI expert, what it could do.
I told him I can’t tell him that. I told him that he knows his profession, his domain. He knows what he would do with those notes, and he might have ideas for what AI could do with it.
Instead of asking me, I told him to go and test his ideas. Now, of course, there is sensitivity here. I advised him to create a fake patient and to mock up a few notes for them. Something representative of a real patient, but without confidential data, and to then feed that to AI and ask it what he wants to test.
That could be you. That should be you. You know your business and your domain better than anyone else. You know the problems that you face, and you might have ideas for where AI could help. You might not know whether AI could actually do what you want, but you also have the ability to validate it yourself.
That is the core message of this blog post. AI has not removed the need for subject matter experts. For someone who understands the business and the problem, AI has in fact empowered these individuals. Artificial Intelligence needs your Human Intelligence.
Of course, there is a difference between validating an idea and deploying a robust and secure solution to production. The ChatGPT example I showed is nice, but it is nowhere near as polished as Google Illuminate.
While manually copying a research article might be fine for today, you’d of course want to automate the process. Your data is sitting in SAP ECC, or SuccessFactors, or maybe a data warehouse or some other system, so even getting to it might be a challenge.
Once you have access to the data, you have to be careful with its sensitivity. You have probably all heard that you should be careful sending confidential data to the free version of ChatGPT, as their licensing allows them to train future models on your data, but you also have to be careful of your LLM bypassing your security and confidentiality restrictions when answering questions.
To get all of this right, you have to wade through jargon soup. Which model to use? ChatGPT, Claude, Gemini, Llama, something else. What about a platform? AWS Bedrock, SAP’s Generative AI Hub? Techniques and technologies? Fine-tuning, Retrieval Augmented Generation or RAG, Vector Databases and much, much more.
Not even to mention hallucinations.
The good news is this is where we, EPI-USE Labs, can help. Not only have we delivered Gen AI solutions to clients, but our commitment to AI predates Gen AI by years. I’ll briefly mention two example products. The first is Employee Retention Analytics or ERA which sits on top of SuccessFactors and uses predictive AI to allocate probabilities to retention metrics, such as likelihood to retain per employee and the factors that have the largest impact on each employee’s retention likelihood.
The other is BioBolt, which is an on device facial recognition product with built in liveness checks that integrates with SAP Cloud Identity Services. It allows your employees to log into SuccessFactors, or other SAP cloud services, through facial recognition, on a computer, mobile or tablet, while protecting against bypass attempts like showing a photo.
If we turn our attention back to Generative AI, I’d like to mention two projects we delivered for our clients.
The first is for one of the foremost universities in the United States. They provided a monthly report pack to their faculty members which contained data on everything they could possibly want to know. This included details such as enrolment numbers, fiscal details, grants, trends, student feedback, and lots and lots of others. These report packs were dozens of pages long and dense, and were routinely ignored.
Through our partnership with Snowflake, we designed a data warehouse strategy for the university and then created an AI agent which allows faculty members to ask the specific questions they want answers to, rather than wading through these report packs. Interestingly, the AI itself doesn’t find the answer. Rather it generates a structured query which is run against the Snowflake data warehouse to retrieve the appropriate data, before generating a natural response.
This provides several benefits, especially in terms of maintaining data sensitivity restrictions and avoiding hallucinations.
The other example is a project for a major architectural and heavy machinery client. Think tractors and harvesters and those type of things. These machines are a lot more complex to service and repair than the car you might be driving, and our client struggled with their mechanics misdiagnosing problems or applying the wrong repairs, despite their training investment.
In partnership with Data Robot, we created an AI agent for them which can help diagnose and suggest repairs to the problems encountered by these mechanics. The AI agent was trained on a combination of their service manuals, training guides, knowledge bases, historical tickets and reports, and a bunch of other data sources. The end result is a chat interface where mechanics can provide symptoms, error codes, and observations and receive instructions on how to repair these. Quite importantly, we also provide “citations” or links back to the original service manual or training guide which describes the problem and fix.
Artificial Intelligence is here and it is ready to be used. It might seem like ChatGPT came out of nowhere, but it was the culmination of decades of research and builds heavily on prior techniques. The tech is impressive and capable of much more than what we’re using it for at the moment. What’s missing is you and your ideas. Your business understanding.
Artificial Intelligence needs your Human Intelligence.