Artificial Intelligence: UK policy and global concerns.
A recent document described UK policy on the use of artificial intelligence in the reduction of carbon emissions (BEIS, 2022). This involves co-funding a ‘virtual centre of excellence on AI innovation and decarbonisation’ and funding projects ‘which further the development of AI technologies to support decarbonisation’. Comment on the document noted the increasing use of AI tools to ‘solve some of society’s current and future challenges’, and referred to the UK’s National AI Strategy. It also noted that while ‘AI has great potential, it can also cause harm’, and referenced the ‘AI Bill of Rights’ proposed for the USA (E&T, 2022). The two themes of UK AI strategy and global concern over the use of AI will be developed below.
The UK’s National AI Strategy outlined a ‘ten-year plan to make Britain a global AI superpower’ and foresaw a profound impact by AI on the UK and the wider world during that period (BEIS 2021). The document referred to earlier initiatives such as the 2017 Industrial Strategy ‘which set out the government’s vision to make the UK a global centre for AI innovation’ and the 2018 AI Sector Deal designed to ‘boost the UK’s global position as a leader in developing AI technologies.’ The National AI Strategy set out to build on existing strengths and recognised the power of AI to ‘increase resilience, productivity, growth and innovation across the private and public sectors.’
The
UK Government is advised on such matters by the AI Council, set up in 2019. It
describes itself as focussing on three main areas: developing the public
understanding of AI, tackling negative perceptions and boosting confidence in
AI among businesses and society; increasing skills in AI and the diversity of
people working in the field AI; and ‘exploring how to develop and deploy safe,
fair, legal and ethical data-sharing frameworks’ (AI Council, 2023).
The
AI Roadmap published by the AI Council lists sixteen ‘recommendations to help
the government develop a UK National AI Strategy’. These include two topics
related to carbon emissions, and the document also has a short section on
climate change (AI Roadmap, 2021). Carbon emissions and AI are mentioned in
relation to developing smart materials for energy storage, and in the general
sense of access to data, governance, and the development of cleaner systems,
products and services. The section on climate change contains a number of
general statements on the essential part that AI technologies can play in areas
such as reducing greenhouse gases in the atmosphere and ‘the environmental
impacts of goods, services and human activities’ and mentions more specific
applications in forecasting supply and demand in real time, combating illegal
deforestation and understanding Arctic sea ice loss. However it is perhaps fair
to say that much of the section is merely aspirational. Elsewhere the document
addresses the place of AI in areas such as supporting high growth AI startups,
enabling public sector adoption, health and social care, and defence and
security. It also notes public scepticism towards AI, and the need for a
stronger governing environment: there is an emphasis on reassuring the public ‘that
the use of AI is safe, secure, fair, ethical and duly overseen by independent
entities’ and on giving confidence to business. Some readers may feel that
public scepticism towards AI is seen primarily as an unfortunate obstacle in
the way of economic opportunity, rather than an expression of well-grounded
fears.
A
rather different attitude to the dangers of AI is shown by the Blueprint for an
AI Bill of Rights, which opens with the statement that ‘among the great
challenges posed to democracy today is the use of technology, data, and
automated systems in ways that threaten the rights of the American public.’
There follow examples in which automated systems can be unsafe, biased,
discriminatory, or undermining of privacy (White House, 2022). Five principles
are identified ‘that should guide the design, use, and deployment of automated
systems’. They are protection from unsafe or ineffective systems; protection
from discrimination by algorithms and systems; protection from abusive data
practices; notice that an automated system is being used and an explanation of how
this may affect outcomes; and the option not to use the automated systems but
to have access to a human alternative where appropriate. The discussion of the
five principles is supported by examples, possible actions to implement good
practice, and a range of citations.
A
further approach to the future of AI is outlined in an open letter published by
the Association for Mathematical Consciousness Science which addresses the
question of consciousness in AI systems (AMCS, 2023). The letter describes
itself as ‘a wakeup call for the tech sector, the scientific community and
society in general to take seriously the need to accelerate research in the
field of consciousness science’ since the ‘increasing computing power and
capabilities of the new AI systems are accelerating at a pace that far exceeds
our progress in understanding their capabilities and their “alignment” with
human values.’ The authors of the letter cite the Large Language Models such as
ChatGPT and Bard, and consider it inevitable that such systems will soon ‘be
constructed to reproduce aspects of higher-level brain architecture and
functioning’, and may possibly go on to have ‘feelings and even human-level
consciousness.’ They note that AI systems ‘have already been observed to
exhibit unanticipated emergent properties’ and their future development may
change ‘what society can do to control, align and use such systems.’ (Emergent
properties in the context of the future of AI form part of an online discussion
(AI Revolution, 2023) and in the context of Large Language Models they are the
subject of a paper by O’Connor, 2023).The emergence of consciousness in AI
systems would raise ethical, legal, and political concerns beyond those
presently recognised, and the writers call for accelerated research in
consciousness ‘in order to ensure that AI development delivers positive
outcomes for humanity’. Reference is made to a previous open letter, calling
for an immediate pause of at least 6 months in the training of AI systems more
powerful than the language model created by OpenAI and numbered GPT-4 (FoL,
2023).
Problems
increase as AI systems approach Artificial General Intelligence, where they
equal or surpass human capabilities in most tasks, or go beyond this to Artificial
Super Intelligence (ASI), possibly becoming self-aware and significantly
surpassing human intelligence. Tzimas (2021) speculates that ‘ASI systems could
very well decide to restructure our political and social systems, eventually
even suppressing humans, with the best of intentions, as for example to reverse
climate change or secure human welfare.’ Benekova et al. (2022) warns that ‘we should by no means be
looking for an AI that is too similar to human intelligence’, given that ‘science
has created or contributed to causing a number of global problems that are
not just a
menace but even
present a real
existential threat to
mankind, such as weapons of mass destruction or climate change’.
In the context of climate change, even if we limit discussion to present AI systems, it is not difficult to see how just one of the ills to which AI systems are prone, namely unconscious bias, could affect the decisions made by them. Such bias could be on the grounds of race, religion or politics, favouring one group over another on matters such as the allocation of resources. At a broader level, bias could exist regarding species, affecting decisions on whether or not to provide protection to a given organism.
References
AI Council, 2023, AI Council, online, accessed 9 May 2023
https://www.gov.uk/government/groups/ai-council
AI Revolution, 2023, The AI revolution: Google’s developers on the future
of artificial intelligence: https://www.youtube.com/watch?v=880TBXMuzmk
AI Roadmap, 2021, UK AI Council, online, accessed 9 May 2023
AMCS, 2023, The Responsible Development of AI Agenda Needs to Include
Consciousness Research, Association for Mathematical Consciousness Science,
online, accessed 10 May 2023
https://amcs-community.org/open-letters/
BEIS, 2021, National AI Strategy, online, accessed 8 May 2023
https://www.gov.uk/government/publications/national-ai-strategy
BEIS, 2022, Government launches £1.5 million AI programme for reducing
carbon emissions, BEIS, press release, online, accessed 8 May 2023
Benekova, Z., et al., 2022, Artificial Intelligence, Value Alignment and
Rationality, Sciendo, online, accessed 8 May 2023
https://sciendo.com/article/10.2478/bjes-2022-0004
E&T, 2022, UK launches £1.5m AI green innovation programme, Engineering and Technology, Nov. 2022, online,
accessed 8 May 2023
https://eandt.theiet.org/content/articles/2022/11/uk-launches-15m-ai-green-innovation-programme/
FoL, 2023, Pause Giant AI Experiments: An Open Letter, Future of Life
institute, online, accessed 10 May 2023
https://futureoflife.org/open-letter/pause-giant-ai-experiments/
O’Connor, 2023, Emergent Abilities of Large Language Models, AssemblyAI, online,
accessed 10 May 2023
https://www.assemblyai.com/blog/emergent-abilities-of-large-language-models/
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3947615
White House, 2022, Blueprint for an AI Bill of Rights, online, accessed 8
May 2023, online, accessed 8 May 2023
https://www.whitehouse.gov/ostp/ai-bill-of-rights/
Comments
Post a Comment