It’s amazing how far we have come. Looking only 18-24 months back, one can see the huge progress (together with a lot of buzz, admittedly) AI technology has made. And there is a lot more to come.
Generative AI brought a fresh air to the technology landscape, revolutionizing not only what can be done, but also the requirements to accomplish it. One clear example is copilots for programming: one can ask an assistant to create a complex code for solving a certain problem, sometimes even without actually knowing anything about programming! Likewise, one can create the lyrics of a song with ChatGPT and then use them with apps like suno.ai to generate the tune. All together in less than a minute. Recently, Google launched NotebookLLM which, among other functionalities, allows you to automatically create podcasts from your documents. Astonishing.
In the AI landscape, there are several dimensions that have changed a lot. For instance, building AI models for specific tasks has never been so easy, just because Large Language Models (LLM) have already part of the knowledge needed to solve many tasks. Hence, if you could take a time machine to travel ten years back and tried to explain to an AI/Data scientist techniques such as prompting, RAG, and the like, you would very likely be considered a visionary… or simply seen as plain nuts.
But, is AI going in the right direction ?
Let me tell you something that happened some days ago. I was sharing on LinkedIn a summary of the talks I was giving on the following few days. Then, apart from the usual suspects that reacted to my post, there was a particular name that caught my attention:
Since my post was in Spanish, I thought it was a bit strange to have some reaction from non-Spanish audience. So I went on to know a bit more about Tess …
According to its LinkedIn profile, Tess is the first fully artificial intelligence employee at its organization. Hence, in spite of the nice picture on its profile, Tess is just 0’s and 1’s, not a real person after all.
Now let’s stop a minute and think about what Tess has done when reacting to my post: Was it one of the tasks this AI is supposed to do in its daily work? Does that activity impact its work? Honestly, I don’t think so. I’m not planning to rent or sell any property, and I really wonder why Tess considered my post to be leading to a lead… Hence, I can only conclude that Tess’ actions are misaligned with their owner’s intentions.
In a recent manifesto written by experts in the field, it is shown how the alignment between the use of AI (assistants) and their intended goals is crucial to guarantee the successful development of the field. This document also foresees an incredible impact of AI in our daily lives for the years to come, and proposes ways to monitor and control AI assistants so that they are continuously aligned and their outcomes are beneficial.
In that direction, companies like Alinia or Galtea are focusing on ensuring LLMs are kept under control when deployed in a software project.
So, yeah, nowadays you can program your AI bots to look around in social networks like LinkedIn and prospect the market you are aiming at. But you should make sure they do not end up focusing on the wrong targets so that your brand gets negative consequences.
Sticky AI
Comparing the big inventions of our history with the current milestones of AI, I think there is still a lot to be done in order to bring AI to the same level of game-changing inventions like electricity or flying. I believe technological progress only impacts society when it is sticky: once you try it, you cannot avoid using it from there on. Electricity is a clear example: it revolutionized the way people lived (for the good), and made the world shift towards a new quality of living levels never considered before. Of course, it is nice that now I can compose music without having to spend time learning it, but in my honest opinion this will not make AI an invention comparable to electricity.
But, what’s sticky for AI ? I cannot agree more with Joanna Maciejewska, who recently said: “I want AI to do my laundry and dishes so that I can do art and writing, not for AI to do my art and writing so that I can do my laundry and dishes”.
The first and unique goal should be building AI technology focused on solving real problems that bring a true impact to society. That will ensure that AI takes a sticky role in our lives. Think for instance in health applications: an AI that may shorten considerably the time needed to predict a disease would definitely be adopted without hesitation.
But as Joanna suggests, for its sudden adoption, the tasks AI is solving don’t need to be critical. Digital productivity is an example of this: we currently spend one third of our working day on our email. On average per day, we have around 58 interruptions and use between 10-15 apps. This causes our daily interaction with computers a real challenge as we try to keep mundane and tedious tasks under control so that we can do impactful and creative tasks that really bring value to our organization. AI apps dedicated to improve our digital productivity may be a good example of how sticky AI can be.
What may be some indicators that an AI solution will be sticky for a particular task X? Let me outline some:
- The current way of solving X takes too much, or
- The current way of solving X costs too much (money, lives,…), or
- Task X is crucial for higher-level task Y which falls in some of the two categories above, or
- The new way of solving X creates an unprecedented need (e.g., before commercial flights, travel was considered a luxury, afterwards, a commodity)
Of course, the previous list is not exhaustive but rather the result of analyzing the success of some inventions that came to my mind when writing this article. There may be other indicators that may make an AI sticky.
In conclusion, one of the key risks of developing non-sticky AI technology is the potential for it to become irrelevant. When AI is used to replicate human behavior in tasks where we don’t particularly excel—such as liking posts on LinkedIn—it risks being sidelined. However, if AI is harnessed to address real-world problems in innovative, non-human-centric ways that lead to impactful and hard-to-ignore solutions, it is far more likely to have lasting significance and adoption.
Josep Carmona
Process Talks & Catedra.ai
