Imagine this scenario. You want to order your shopping for the week, but at the same time are bored with your usual recipes. Common story, right? Imagine further that you call on an AI tool to help you here.
Your requirements would be to suggest 7 meals very different from one another and then to order a list of shopping items to be delivered to your home address tomorrow lunchtime. Further still, you want each of the meals to contain five of your ‘5 a Day’, to add up to £60 maximum, to contain no nuts whatsoever as you’re allergic to them. Oh, and you don’t need any salt or cooking oils because you’re well stocked up with these items already. Finally, you’d like these to come from your local Aldi. For simplicity, let’s suppose you’re ok with providing your credit card, location and login details to the Aldi website.
Now imagine it’s conversational. Therefore, rather than specifying all those ‘Ands and But’ scenarios, you don’t need to prebake a loaded prompt into the AI, but naturally instead it can come back and ask you, ‘Do you have any allergies?’ ‘Where would you like to shop?’
Well if you think about this, doesn’t it feel very close to real life and perhaps something you might speak about with someone living in your house? It’s interacting with an AI bot just as you would with a human, asking the same kinds of questions and getting the same kinds of answers. However, there are added benefits like getting inspired with new dishes, perhaps dishes you never would’ve considered; the AI will have a huge array of potential places to draw inspiration from as opposed to your five or six cookbooks and quick online search. This element of creativity is a true value add to you as a human.
Furthermore, all this would probably be a 2-minute process all the way from your first question to having an order placed, compared with half a day or so. Forcing the AI to keep the budget under £60 may also save you money and stop you getting swayed by merchandising for instance. It’s an extremely-simple process and you can focus on other more interesting things you have on your bucket list.
This is agentic workflows orchestrated by AI, and once it understands the full request it can call on other agents to specialise in their own tasks in sequence. If you think the number of tasks, it’s quite a complex process but it does mimic what a human would think and do. To me, what’s fascinating is how this is what AI is truly about. I don’t mean just helping with shopping, but more generally being able to act like a human to help matters, and maybe even helping us think differently or creatively. It’s something to make your life easier, but to act as an assistance rather than a driver.
Of course there is a risk of it making mistakes such as to present peanuts in one of those dishes, for example; this is why humans are important to supervise and quality check the AI pseudo-decisions, rather than to be overly reliant. Then, where changes are required the process to get a better answer is simple and quick, i.e. ‘This dish has peanuts. Please give me better recipes.’ Humans make mistakes and we often learn to improve via those mistakes, so agentic AI would be doing the same.
While from the face of this it isn’t necessarily a life-changing use case, it absolutely could be for someone who’s wheelchair bound and living alone. Imagine them now being empowered to speak to their bot to complete tasks they would have struggled greatly with before.
Then extrapolating this idea beyond shopping, actually the principles of asking AI to do a bunch of processes would equally apply in cancer research: instead imagine it’s a lab researcher asking domain-specific questions of her bot, and not needing to read through plenty of books, thereby massively expediting the research process and likely leading to more findings. It quickly gets serious when one thinks of AI in healthcare.
Therefore for me, yes ChatGPT + DALL-E have opened all our eyes. Also I’ve worked in AI or its statistical origins for over 20 years and yes machine learning is also vital for organisations. However, I’ve never been more compelled by the potential of agentic – it is the true way we’re going to see how good AI can really be. When I was young, I recall AI being the Terminator, so something which can learn geometrically and act like a human. Of course that analogy is somewhat morbid and we’re far off that. However when focusing on the pluses of such an opportunity, this is far closer than machine learning and even ChatGPT how we all saw it in 2022. I think we all have seen agentic needs much development but the promise and excitement are definitely there.
Hope you found my take interesting. Below are the key takeaways:
- Inspiring recipes listing the necessary shopping demonstrates agentic AI
- It’s capable of complex behaviour akin to real life, with potentially new benefits, e.g., creativity, efficiency
- Widely applicable, e.g., medicine, flying a plane
- Reliable assistant for a human. (Human should check quality)
- It’s the main promise of AI, finally. Everything before feels a build up
