I want to reflect with you on the trends and the future of AI, focusing specifically on the near future. Although it is quite a daunting task, a few years ago, I would not have expected 2023 to bring such rapid development in this field. The scale of changes is admirable on one hand, but on the other, it compels us to ask ourselves difficult questions. A moment of reflection can help make both the questions and the answers more valuable.
Update from 20th November: The recent events in the world of AI clearly demonstrate how much is happening. Sam Altman, the creator of ChatGPT, was reportedly fired from OpenAI, and the entire story surrounding this event is being considered for a film adaptation. Currently, Sam Altman has received a proposal to build a similar team at Microsoft, with many OpenAI employees expressing their willingness to follow him.
The Third Phase of IT Development
To start, I would like to refer to the division mentioned by Dharmesh Shah. We can distinguish three phases in the development of information technologies:
The phases mentioned above indicate the potential that accompanies them. The current phase of human collaboration with AI is supposed to bring unprecedented cognitive abilities, which would enable us not only to access information but also to better understand it and – in a sense – to experience it. This, in turn, may change our thinking about education and contact with information in general.
Gartner points out another interesting aspect – in the era of AI, technologies cease to be just tools, and become teammates. We once thought that such a change would be associated with robotization. To some extent, this is still relevant, but a “teammate” can take an immaterial form. This is visible in how we often formulate prompts. Our thinking is changing from what a machine can do, to what (or who!) it can be.
Let’s go a step further. The fact that AI can support humans is just the tip of the iceberg. Artificial intelligence can “communicate” much better and more efficiently with a computer or another artificial intelligence, if only because it is not based on natural language, but can use more optimal methods. Moreover, we are already witnessing the efficiency of training artificial intelligence with content generated by… artificial intelligence itself.
I assume that many future models will be based on the achievements of large companies that have, in a sense, paved the way. At the same time, the cost of creating these models is already incomparably lower. As a result, they can be more specialized and accessible.
What engineers have to deal with is so-called hallucinations. Large language models (LLMs), which include ChatGPT, are such yes-men. They often claim to know the answer, though in reality, they sometimes fabricate it. In ChatGPT, you can set an additional prompt that is added to each query – I set it so that ChatGPT always adds a note on how sure it is of its answer. It’s always sure, although sometimes I see it writes nonsense! And I use the latest, paid version.
By eliminating hallucinations and thus increasing the credibility of artificial intelligence outputs, our satisfaction with working with it will not only increase. Think about the potential this will give to artificial intelligences working with each other…
There are many dangers associated with artificial intelligence, and I plan to dedicate a separate article to the strict issue of AI safety.
Transition on Three Levels
What is also available, although based on my experiments works moderately, is the transition from a static to a dynamic model of AI operation. What I mean by this:
- In the static model, we get answers from a network that is once taught and possibly re-taught;
- In a dynamic approach, AI uses current internet resources (or any other set of data from which the current version of AI will use).
Improvements will occur in both models, but those that can provide answers based on current information will be especially useful. This will give incredible potential in combination with process automation (RPA, Robotic Process Automation).
At the same time, we see another transition – what a prompt really is. It’s no longer just text, but various information carriers: photos, spreadsheets, programming code, databases, etc. The future is more interactive use of AI capabilities – voice conversations (also already available), eye tracking, and maybe even a brain-computer interface (Elon Musk initiated the Neuralink project).
The third change – if we look at AI as a broad group of solutions – may rightly be considered an illusion. I mean the transition from AI working passively to actively. After all, artificial intelligence operating in the background (actively) supports the operation of Netflix suggesting movies to us or TikTok, which is supposed to keep us as long as possible, thanks to the perfectly matched sweetness of kittens and puppies (or whatever we like ;)).
However, generative artificial intelligence works rather passively – we have to manually enter a prompt. I believe that this will change and AI will act even more actively to a greater extent. It will predict what we need, and also “handle” certain matters for us in the background.
Let’s stop at the concept resulting from the aforementioned third transition: from a passive version of AI to an active one. The future may involve artificial intelligence personalized not for a specific application, but for a specific user and his needs. Such AI could be compared to a personal assistant with vast knowledge and… patience. We could entrust it with the most tedious and repetitive tasks, but also count on very personalized answers. The future may be the time of AI agents.
In the most likely form, it may be an even more advanced Google Assistant, Apple’s Siri, or Amazon Alexa. We have already heard a recording from Google where the assistant makes an appointment on our behalf and sounds just like a human. The process of creating integrated chips that ensure efficient operation of neural networks directly on our phone is noteworthy. Thus, this personal AI agent would not have to transmit any information about us to the cloud.
What I write here may cause a smile of pity: perhaps conversations de facto between artificial intelligences will be popular. We write an email using AI, and then the recipient’s AI develops its summary and suggests a response. We just click send.
“The Third Benefits”? Competition Between AIs
What, apart from the application, most distinguishes the various models and what for now will not change, are the data with which they are powered, as well as performance indicators: the number of parameters, computing power, and the degree of optimization. These are not very marketable parameters, and OpenAI, after the huge sensation that ChatGPT made in the free version, no longer arouses such emotions. I assume, therefore, that it will be a time of competing for who will implement which model into their applications and do it better. AI will therefore be largely an added value to existing solutions.
And here we come to perhaps the most certain prediction. I assume that artificial intelligence will merge with our lives and will be a component (often key) of solutions that we know and will get to know. Current solutions, such as ChatGPT or Midjourney, will gain new interfaces – their own “wrappers”, as well as in the form of third-party applications. This is already visible, but the real change will begin when AI is not a buzzword, but an obvious functionality.
Three specific examples:
- Microsoft with its Copilot integrated with Windows and Office;
- Adobe, which thanks to Firefly integrated with Creative Cloud applications will change the way of creating forever;
- HubSpot, which thanks to data analysis affects marketing and sales.
Soon, perhaps even deeper changes will affect search engines. For now, we see a chat feature integrated with Bing. But maybe Apple or Google will offer their intelligent search engines, which will cover both our data (after all, we have a piece of our life on the phone) and the internet, and by knowing us, will deliver even more accurate results. (The fact that Google already uses artificial intelligence to rank search results is obvious to me. Every click of ours and staying or leaving the page teaches the algorithm.)
In my opinion, the coming months will be a time of certain disappointment and withdrawal – we are becoming increasingly aware of the current limitations of artificial intelligence. However, it is worth keeping up to date and developing competencies related to AI (companies, I’m also talking to you!).
Three words at the end
In a separate post, I wrote a bit about the job market in the era of AI. About 3 months have passed; at the current pace, that’s a lot, but the message of that article remains current. I encourage you to read it.
We have interesting months and years ahead of us that will amaze and frighten. I hope this article has helped you get a better grasp of the trends that are emerging.
What will be, for example, in 2033? We can speculate, but with a high probability, we will be wrong. Probably just as we shaped AI, now AI will shape us, to a limited extent of course.
I invite you to share your thoughts in the comments. I also encourage you to subscribe to the newsletter, because AI will appear on the pages of my blog more than once.