As 2022 drew to a close, once again the overblown hype surrounding fully automated driverless trucking fleets and ubiquitous cross-country journeys in autonomous vehicles—which if Elon Musk and his like are to be believed, should already be here—is exposed as one more chimera in the mist. Far from causing mass unemployment in the professional driving fraternity, the world is currently facing a shortage of drivers of all forms of commercial vehicles, from taxis and buses to road- and rail-based trains. Even the aviation sector—long supported by the software-based autopilot tool—is facing unprecedented human capital deficits.

However, as 2023 dawns, it is becoming increasingly evident that the fastest-moving—and possibly most socially and economically significant—new kid on the artificial intelligence (AI) block is the family of natural language processing tools. These tools bring together the combined knowledge of linguistics and mathematical algorithms to take a real-world language input (either spoken or written), process it, and make sense of it in ways a computer can understand. The tools can then create a vast array of outputs—oral and written—in response to input prompts.

While early iterations of the technology gave us tools such as Google Translate, which takes a string of inputs in one language and gives outputs in another, more recent variants, such as Bidirectional Encoder Representations from Transformers (known as BERT), Generative Pre-trained Transformer 2 (known as GPT-2), and Generative Pre-trained Transformer 3 (known as GPT-3), provide far more creative output. These tools are designed to search for and provide answers to direct questions (such as those that might be asked of a chatbot), complete answers to essay questions, or even poems and songs in whatever style the requester asks. Using similar processes, the algorithms are also able to create visual art outputs.

My pick is that the hot topic for AI and machine learning in 2023 will be the relationship between these new AI tools and creativity. 

On the one hand, AI artists claim that the works generated are novel and attributable to them because they crafted the prompts that led the software to create the output in question. On the other hand, I have a problem with accepting and grading a student’s essay generated using these tools. (And I have little doubt that these have already been presented to me for assessment. The tools are very sophisticated and it is hard to distinguish “fakes” from genuine efforts; though ironically, the absence of spelling, punctuation, and grammar errors may be a big clue for some students). Yet other educators have opined that if the tools allow students who would otherwise face limitations in producing outputs to create something of which they can be proud, then that is sufficient to encourage their use by those students. Others suggest there is nothing wrong with using AI-created exemplars to stimulate further creativity if that gives the student confidence to proceed along that path.

These issues, however, seem to merely start scraping the surface of the questions to be addressed. Should newspapers and websites be allowed to publish AI-created stories or attribute them to named journalists? And how would we know if they had used AI-generated content? As the tools can create research papers that encompass variations on existing themes and are sufficiently plausible to convince human peer reviewers of their novelty, should they be published in reputable journals? If so, who is attributed with the research credit—the AI tool or the researcher creating the prompt? And how should this research be valued (or even identified) compared to human research—which even now is aided by the use of search engines for important components such as compiling literature reviews? For example, how should a paper be evaluated if the researcher consigned the literature review to the AI tools, but then conducted and wrote up a novel experiment without electronic assistance?

I, for one, look forward to seeing where this debate proceeds in 2023.

The post Artificial Intelligence and Machine Learning in 2023 appeared first on American Enterprise Institute – AEI.