What can we expect from AI and Machine Learning in 2023?

In 2022 we saw extensive growth in the application of artificial intelligence (AI) and machine learning (ML), with the adoption of both technologies doubling and the proportion of organisations using AI/ML drifting between 50 and 60 percent since 2017.

Therefore, it is no surprise that the growth of AI and ML has played a significant role across an array of sectors, including data collection. The importance of data collection, in particular web scraping and alternative data, cannot be overstated in the current world, as having access to a wide range of information is crucial for businesses looking to stay competitive.

AI and ML are already profoundly impacting society and will continue to do so in the coming years. So, with much emphasis and reliance on AI and ML, what can we expect in terms of their development in 2023?

The emergence of unseen machine-learning capabilities

AI/ML technologies host an array of previously unseen capabilities. Growth in large language models is expected, as well as in self-supervised machine learning methods such as contrastive learning. Large language models for NLP (like BERT, GPT, and their derivatives) will keep improving, and their use will become more pervasive. One pretrained model will be able to be used with little modification for many functions, including sentiment analysis, summarisation and word sense disambiguation.

Also, it is anticipated that more will focus their attention on contrastive learning, aiming to learn representations of data by contrasting between similar and dissimilar samples. Transformers, text-to-image, and diffusion models require large-scale datasets and supervised pre-training of large models is extremely expensive. Therefore, self-supervised contrastive learning can be used to leverage vast amounts of unlabelled data to efficiently pre-train large models.

Furthermore, contrastive search is a related technique which significantly improves the output of large language models when used for text generation tasks.

Content generation techniques to become profitable products

In 2023 we can expect to see an increase in content-generation techniques. There is a likelihood of the continued evolution of Stable Diffusion, GPT-3, GitHub Copilot, and other content generation techniques into profitable products used by developers and content creators in real-world applications. We may see an increased interest in multi-modal models that can handle text, images, audio, and other inputs for multiple tasks.

This will lead to a shift from using AI for static tasks like classification to language-model-driven interactive workflows that help people perform tasks more efficiently.

Adoption of AI within the biotech industry has also been on the rise in recent years, helping improve the speed and accuracy of drug development, and benefiting patients and the healthcare system. Overall, AI adoption within biotech will keep accelerating, benefiting not only drug discovery efforts but our general understanding of cell biology.

AI apps to take over but regulations to be increased

As AI technology continues to advance, it would not be a surprise if ChatGPT replaced Google in some ways and OpenAI emerging as a big tech giant on top of this product. But one must be aware of its impact on education, healthcare, and personalised software and how it may transform our society in many ways.

Various regulations for AI-powered tools that emerged in Europe, US, UK, and Canada in 2022 will continue in the coming year with stricter regulations and their implementations. Additionally, there are now lawsuits emerging for the creators of specific models, claiming that the data used for training was subject to copyright and was acquired without the necessary precautions.

Subsequently, we will see a proliferation of apps built on top of AI-generated content (text and images) through tools like Dall-E and Stable Diffusion, which may impact open-sourcing stable diffusion on the AI community. It’s no secret that the increased interest in AI and NLP in bots has brought great results.

For instance, OpenAI - a research institute and technology company that is focused on advancing the field of AI, created the well-known ChatGPT as one of their projects. The

technology allows bots to understand and respond to human language, providing more natural and human-like interactions.

While Large Language Models, the technology ChatGPT is based upon, currently can produce outputs that are often called “confidently wrong”, these issues can be mitigated in the near future. Most machine learning models can also provide their confidence rating, which provides a prediction of how correct the output may be.

More use cases for AI-powered applications

Lastly, we should expect to see more AI applications and the adoption of generative AI when moving in 2023. In general, it seems that the year will be largely defined by the proliferation of generative AI unless legislation or under circumstances put a stop to the production of such technology.

The introduction of technologies such as GitHub Copilot, being the first of their kind, has opened up the world of AI/ML leading the way for improvements and the introduction of related technologies in 2023. Furthermore, tools such as Lensa, which uses Stable Diffusion to create photos and content for social media, are paving the way for authors and illustrators to integrate AI/ML tools into their work.

Development and programming may also be heavily influenced by generative AI. Programming languages are syntactical, meaning that there is always only a single definition and interpretation. Natural language is significantly more complex due to the inherent ambiguity it possesses, which means that generative AI should, in theory, have a harder time becoming equal to humans in writing. Programming, on the other hand, has less (or none) ambiguity, making it easier for generative AI to solve, which should allow it to produce more accurate results faster.

Yet, replacements are due to happen soon. Both writing and programming will require humans to evaluate the output, which will require expertise to understand whether the generated content is of high quality. Generative AI will likely be an extension rather than a replacement for humans.