December 22, 2022

AI trends to watch in 2023

AI trends to watch in 2023

It is always tricky to predict which technologies will have the biggest impact in the coming year. However, in 2022 it seems pretty obvious. Generative AI visibly stood out and we will for sure remember this year as the year that Generative AI stunned the world with its impressive capabilities.

What does the future hold for Artificial Intelligence? We asked our Senior Machine Learning engineers to shortlist their take on what will happen in the upcoming year. This blog post is a part of the IT Trends series by SoftwareMill. Have a great read and Happy 2023!

Image generated by Midjourney

Our take on AI top trends to watch in 2023

Adam Wawrzyński

In your opinion, what was the top AI/ML event or trend of 2022?

No doubt creative AI. In 2022, we had a rash of models that were able to generate images based on a textual prompt (e.g., DALL-E) or assist in textual problems (ChatGPT). We are at such a stage in the sophistication of these models that it is often difficult to distinguish between human and AI work, as was the case with the Colorado State Fair's annual art competition, which was won by DALL-E 2 with its work "Théâtre D'opéra Spatial."

What is the next big trend in AI/ML for 2023?

I think that in the next year, scientists encouraged by the effects of creative AI will continue research in this direction. From the point of view of ordinary users, I expect many products that will try to use these creative models to support people in their tasks, such as generating images, designing graphical interfaces, writing and analysing code and texts. AI support for the masses!

Kamil Rzechowski

In your opinion, what was the top AI/ML event or trend of 2022?

In 2022 there were few astonishing publications that pushed SOTA AI to the next level. Among others STEGO (Unsupervised Semantic Segmentation by Distilling Feature Correspondences) that allows semantic image segmentation with so far unprecedented accuracy. AlphaTensor (Discovering faster matrix multiplication algorithms with reinforcement learning article) that proves that AI can resonate about things differently than humans and find better solutions for old problems.

And finally Dual-Shutter Optical Vibration Sensing that allows to isolate voice/music tracks, simply by looking at the sound sources with a camera. I only picked 3 most interesting papers in my opinion. However, the year 2022 did not bring such breakthroughs like 2021 with StyleGAN v3, GauGan and Bootstrapped Meta-Learning or 2020 with DINO and NeRf. The 2022 was more about pushing limits of existing architectures and solutions.

In 2022 trends were towards creative AI. We saw many AI models creating music, images, poetry, videos and even games. DALL-E 2 allows you to create incredible art from the sentence. Stitch it in Time allows you to swap or modify faces in the video stream with much better realism effect, than before and fewer training data. Make-a-video allows text to video generation, without text-video data. Creativity is often seen as a human skill and seeing it right now quickly growing in the AI field makes it feel more human.

Image source

What is the next big trend in AI/ML for 2023?

I’m really looking forward to quantum AI. I hope that 2023 will finally bring quantum computing closer to commercial applications. However, quantum computing will not substitute the silicon based processors, and regular AI will still be developed. 

I believe that in 2023 we will see unsupervised learning pushing new limits and AI models building more in the Routing Networks fashion to mimic how the human brain works. When we process information, we use only part of our brain. I expect to see the same direction in the AI field: building bigger models that have parts of the network specialised in different tasks.

Maciej Adamiak

In your opinion, what was the top AI/ML event or trend of 2022?

In my opinion, the top IT trend in 2022 is utilising Stable Diffusion (Dall-E) in all possible scenarios and by everyone, from curious machine learning engineers to artists. In addition the appearance of ChatGPT caused a lot of hype and controversy. I am sure that there will be a huge discussion in the near future about the use of this type of machine learning in practice, the sources of data used in training, and copyrights of content available on the Web. We are close to clarification of legal issues related to AI. On the one hand, it's good, but I also have a lot of concerns.

What is the next big trend in AI/ML for 2023?

I am really waiting for quantum computing to gain the same velocity as currently machine learning has. Cannot wait for it to happen, be accessible and used on production. Quantum Machine Learning is an exciting topic which has a chance to perfectly use the base built by the ML / AI community. 

The other thing is probably the emergence of generative ML countermeasures. There is a need to verify the authenticity of content generated by various models. I am sure we are in for a flood of content created by ChatGPT. It is difficult to estimate the quality of content we can encounter. However, it should not be considered equivalent to that generated by humans.

Rafał Pytel

In your opinion, what was the top AI/ML event or trend of 2022?

Generative AI, with the most famous examples of DALL-E-2, Stable Diffusion, and ChatGPT. These models are extremely interesting to try out (quite a lot of them are available as an app online), but they are also transforming creative industries, where people start to specialize in giving the right prompts to generate the most spectacular text image or video. 

What is also worth remarking is that due to the open-source character of models and architectures, the advancement in the field accelerated to a speed unseen before. Stable diffusion models were published during the summer, while 2 months later: other models/devices support was quickly added, large adaptation with a variety of tools (e.g. widespread integration with Hugging Face), models deriving from their success to other domains (e.g. video, 3D imaging), and even more applied fields (e.g. Photoshop plugins). By now I am wondering when AI will start winning video contests or get its first Oscar.

What is the next big trend in AI/ML for 2023?

With the rising concern of the large model environmental impact, I think the need for more effective use of resources could be the theme of the next year. If you check some model descriptions on HuggingFace they even offer information about how many hours/CO2 was used. The solution to reducing required resources could be quantum computing, which is gaining more and more traction every year. 

What is your take on AI trends for 2023?

We're curious what your eye has spotted and what in your opinion we will see more of in the upcoming year. Share your trends in the comments!

Do you want to be up to date with AI/ML and technology trends? Join and get unlimited access to Data Times - a monthly dose of tech news curated by ReasonField Lab and SoftwareMill engineers.