This post is available to download as a PDF here.

****

The appearance in recent times of Open AI’s ChatGPT series and Google’s Bard application has resulted in a spate of articles analysing such developments which reflect views ranging from catastrophic conspiracy theories to incredulous debunking (Foucart 2023).

Many of the concerns about the new AI tools have been expressed by educators who fear that teaching and learning will be damaged by the easy access to the large language models (LLMs) like ChatGPT which can write essays and answer assignment questions in a matter of minutes (Heaven 2023). A letter to The Times signed by leading educators from the state and private sector described AI developments as ‘bewildering’ and announced the launch of a review body of experts to advise schools on which areas are ‘beneficial, and which are damaging’ since ‘we have no confidence that the large digital companies will be capable of regulating themselves in the interests of students, staff and schools’ (Shad  2023).

As part of the attempts to deal with the complex practical problems of AI implementation in educational settings, it is well worth considering the ethical implications of such developments.  In this respect, I suggest that it might be useful, both heuristically and pragmatically, to regard certain AI applications which achieve what Nick Bostrom (2016) calls superintelligence – characterized by ‘intellects that greatly outperform the best current human minds across many very general cognitive domains’ (63) –  as incipient non-human persons. This concept was initially introduced by Peter Singer (1975/2009) in his groundbreaking work on animal liberation which posited the notion of non-human persons, particularly animals such as higher primates, who, arguably, have sufficient sentience to warrant a moral status on a par with humans.

Initially, Singer wanted to argue that certain animals should be brought into the moral community on the grounds that they were sentient beings who could suffer.  In recent work, Singer has examined the status of AI machines and robots in the light of the wider questions of rights and responsibilities surrounding the human/AI interface.  Anticipating contemporary developments, Singer (2023) asks ‘if machines can and do become conscious, will we take their feelings into account?’ This is considered a pertinent question since our treatment of the

only non-human sentient beings we have encountered so far – animals – gives no ground for confidence that we would recognize sentient robots not just as items of property, but as beings with moral standing and interests that deserve consideration.  (382) 

Of course, such claims beg the question of whether any of the current AI applications can be considered as either sentient or conscious.  There is no doubt that they satisfy many of Bostrom’s criteria for speed superintelligence which applies to systems ‘which can do all that a human intellect can do, but much faster’ (2016, 64).  He goes on to contend that ‘artificial intelligence already outperforms human intelligence in many domains’ (ibid, 14) and cites AI wins in chess, Go and other games over human grand masters.

However, critics such as Noam Chomsky (2023) have written about the ‘false promise of ChatGPT’ arguing that – although such applications are ‘marvels of machine learning’ – the science of linguistics and epistemology indicate that ‘they differ profoundly from how humans reason and use language (14).  Similarly, Philip Goff (2023) argues forcefully that ‘ChatGPT can’t think – consciousness is something entirely different to today’s AI (1).  Against this, Bostrom’s recent comments on developments are more favourable to AI sentience.  In a recent interview he claimed a fair degree of sentience for LLMs such as ChatGPT, and went on to say that:

I also think it’s not doing them justice to say they’re simply regurgitating text…They exhibit glimpses of creativity, insight and understanding that are quite impressive and may show the rudiments of reasoning… If an AI showed signs of sentience, it plausibly would have some degree of moral status… This means there would be certain ways of treating it that would be wrong, just as it would be wrong to kick a dog or for medical researchers to perform surgery on a mouse without anesthetizing it.  (Al-Sibai 2023, 1-2)

Since there is little to lose and much to gain, the recognition of AI applications as potential non-human partners may help educators to realize and access – as a number of recent critical commentators have noted (Alhazmi et al 2023; Lancaster 2023) – valuable teaching/learning insights and enhancements of practice.  After all, humans have improved at the fiendishly complex game of Go since DeepMind’s AlphaGo finally defeated the world’s best players (Rosenblum 2023) and, on a more serious level, recent research on AI implementations at all levels of education indicate a wide range of positive advantages for teachers and learners (Chen 2023).

In terms of exercising the vital control and regulation of any potential partnerships, Bostrom’s ‘common good principle’ insisting that ‘superintelligence should be developed only for the benefit of all humanity and in the service of widely shared ethical ideals’ (2016, 312) would provide a fitting educational mission statement.