A Year in AI

Over the past year, progress in generative AI continued at an unprecedented rate (Perlow, 2023; Perrigo, 2023-a). Thanks to the progress of Natural Language Processing (NLP), chatbot conversations have become more fluid as bots better understand language and provide quicker, more relevant responses (PM, 2023).

In addition, conversation with chatbots has become richer due to the development of multimodality  (Perrigo, 2023-b).  Originally (It’s worth remembering we’re talking weeks and months, not years), when we had a request, we provided text and got text responses. The use of voice for searching has now become popular, with fifty percent of mobile users using it regularly (Haan, 2023). A countless number of people have made asking Siri or Google Assistant a part of their everyday lives. ChatGPT (GPT-4) has advanced so it can now respond with voice (Metz et al., 2023).

More recently, the technology has advanced so that other modalities such as images and video can also be used to prompt a text response. This year, we saw that in OpenAI’s GPT-4 and then in Google’s Gemini. In a related development, it became possible to use text to generate images and even video. (Perrigo, 2023-c). Users can not only use text to have videos created, but also edit existing videos with text prompts.

Of course, the impact of this progress has been felt beyond the academic world. The growth of AI use in business has been notable. In a survey of the use of generative AI in corporations, one-third of those surveyed are using it regularly for at least one business function (McKinsey and Company, 2023).  What was once a topic for IT departments only has reached management leadership, twenty-five percent of whom are using it for their jobs. They are also looking to the future.  Forty percent plan to increase their investment in AI. Considerations of reskilling and reducing employees need to be addressed. There is work to be done and issues to resolve, but AI has landed in the corporate world.

So, too, has there been progress with values. Concern about inappropriate responses led to the development of something called “constitutional ai” (Mohan, 2023).  A set of principles is developed, and bots are trained based on those, thus reducing the need for external human intervention and bias. The question, of course, remains about what values. Anthropic has made an effort to democratize this process by collecting values from a cross-section of the population (Perrigo, 2023-d). It is recognized as a step in the right direction, not a solution.

There’s one other development this year that may broaden the impact of AI. Open source AI models are now available, allowing anyone to develop their own custom chatbots based on their own choice of data. Meta (Meta, 2024), for example, is currently offering a choice of open-source models focusing on different areas including language, reasoning, and computer vision.

We’ve come so far in a short period of time. People may be surprised to learn that AI actually started its evolution in the last century. For an understanding of how it developed over time, there is a new Special Section containing a brief history. The question now, for ourselves and our students — where will this take us?