6 AI Trends To Watch Out For In 2023
Artificial Intelligence technology continues to develop at an impressive pace – and already 2023 has been a landmark year. So what has happened – and what else can we expect this year?
ChatGPT blows our minds
OpenAI’s ChatGPT, an artificial intelligence-powered chatbot, has become a worldwide smash. The awesome power of AI algorithms has brought artificial intelligence to mainstream attention with a mix of fascination and horror. The accuracy of the text created by ChatGPT is seriously impressive, leading some to suggest that the chatbot could put people out of work.
However, ChatGPT is not infallible. Built using crowd-sourced datasets the chatbot has been known to make factual errors; as an automated system, it has no way to tell if the answers it supplies are actually true or not. However, OpenAI continues to refine and improve the system using human moderation throughout the model training process. We will definitely be hearing more about this technology in future.
AI regulations are coming
Regulators are increasingly concerned about the ‘black box’ nature of artificial intelligence – how do these systems actually make decisions? This lack of transparency raises questions about fairness, equality and bias.
The proposed European Union Artificial Intelligence Act will require businesses to audit and report on their AI algorithms. The US government is formulating its own legislative response to AI while some cities and states are implementing bans on certain types of technology (like facial recognition) which they regard as incompatible with certain constitutional rights.
The new legislation will increase oversight and accountability, helping to reduce bias and costly mistakes caused by incorrect AI decision-making. Often the creators of AI models are not 100% sure how these decisions are made. During 2023 we expect to see a secondary market for model ‘explainability’ developing alongside AI itself. Tools like Seldon Alibi Explain will help operators meet their compliance obligations by showing them the inside of their ‘black box’ algorithms.
It is likely that increased legislation will delay roll-outs of new AI technology initially. However, new regulatory frameworks will also encourage developers to build more accurate systems that create even greater value for users and customers – and help to build trust in a technology that many people still view as ‘creepy’.
AI becomes more pervasive
Artificial intelligence capabilities are being added to virtually every data-driven application to improve features and functionality. AI capabilities are now commonplace, even in the latest smartphones.
Features like image recognition, text extraction, translation and more will continue to improve in speed and accuracy. As AI algorithms become commoditised, we will see these functions added to other apps and services – and like smartphones, much of this functionality will be completely transparent to the end user. With open-source AI models available to anyone, we can expect to see AI appearing almost everywhere.
AI solves data challenges
Where previous efforts have focused on using artificial intelligence to analyse data, 2023 will see AI being applied to the challenge of improving data. Quality of data is essential to delivering accurate, actionable insights – and AI can help to correct errors, remove duplicate records and even reduce data vulnerability.
AI will be particularly valuable as large datasets are transformed for use in different applications. Automatically rejecting inaccurate data during the ETL (extract, transform, load) stage of migrations will help to accelerate deployment and reduce costs because the data is ‘clean’, requiring less housekeeping over the long term.
AI algorithms become more complex
Applying algorithms to analyse a single factor is relatively straightforward, but also quite limiting. During 2023 we expect to see an explosion in the use of multimodal AI algorithms, designed to accept many different inputs simultaneously.
Consider traditional image recognition apps that can analyse a photo to distinguish between a human and an animal; this is an example of a single-mode algorithm. Now consider an app that can distinguish between a person and an animal and read any text displayed within the same image; by analysing people and text, this is a multimodal algorithm.
Tools like Unitary.ai are rolling out multimodal algorithms that can accept multiple simultaneous inputs (video, audio, text, images etc) to analyse user-submitted files, and then detect and flag suspicious content. Multimodal algorithms analyse each of these elements in context, providing greater speed and accuracy of detection. Given that digital transformation projects are heavily reliant on contextual analytics, multimodal AI algorithms will become the norm.
AI will not end the human race
As mentioned earlier, some commentators are warning that artificial intelligence may supplant entire industries, taking thousands of jobs with them. But the truth is that AI is nowhere near capable enough to achieve such a feat.
ChatGPT is very clever at creating accurate-sounding text – but it is not infallible. Every text prepared by the chatbot must be manually reviewed for accuracy. The same is true of any AI implementation, from manufacturing to medicine – the algorithm may be able to prepare accurate recommendations, but it has to be double-checked by an experienced human who can assess context and nuance too.
As such, AI is best seen as a tool to help humans be more productive rather than an outright replacement.
Artificial Intelligence is undoubtedly exciting – as is the apparently limitless scope for its use in 2023 and beyond. Although we have tried to outline and explain some of the trends we expect to see this year, is it also extremely likely that something new and surprising will emerge that very few of us expect. And that uncertainty just adds to the anticipation.