I have wanted to hold this debate for a very long time, and I feel that I need to start by recognising that artificial intelligence is here. I am not going to stand here and deny that, or say that I want to ban AI. I use AI—my Pixel phone has it built in. I speak to that AI, and I let it choose the music that plays when I am driving. AI is here and it is a tool like any other, and I am broadly fine with that.
However, while I hear a lot in the chamber, and in the media, about AI’s potential to support business, to help with healthcare and to increase productivity, I do not hear an awful lot about the risks. Legislators around the world must decide, and decide fast, how far we are comfortable with AI reaching into particular areas, and what limits need to be placed on it. We should keep in mind that AI is learning from humans, and that includes scraping the depths of the internet for forum comments to learn how to act like a human. If society is racist, AI may be racist, and if we value men over women, AI will learn to do the same.
In order to decide on AI’s limits, we have a duty to understand its capabilities, risks and dangers. We must then tell those who are currently developing AI models where they cannot take their technology, before it is too late.
In my region of the Highlands and Islands, perhaps more than anywhere else in the country, we are encouraged to make use of new technologies, to work remotely and to work smarter, and that requires using AI as an emerging technology. For so many of my constituents, including my own employees, the online world is their workplace, and they—like anyone else—have a right to safety in that workplace. Without regulation, however, AI is not safe.
I will speak to a couple of existential threats. The first is the threat to the workforce. As the use of AI grows in industry, jobs are at risk. We need to get serious about minimum income guarantees, and job and productivity opportunities for humans. Companies will automate if it reduces cost, and the conversation on workers’ rights needs to extend to the implications of AI.
We have not only to answer questions around which jobs should be undertaken by humans—perhaps, as a fairly unpopular “Doctor Who” episode suggested, by setting quotas on the percentage of the organic workforce—but to have conversations about the rights of workers to have a say on decisions that have been delegated to AI. That includes protection from liability and the right to have their accident-at-work claim, or their disciplinary—or even dismissal—case, heard by other humans.
Secondly, the climate impacts of AI are already massive. Nuclear energy business is booming as tech giants buy up reactors on a huge scale in an effort to cut down their emissions. Those emissions are soaring as servers support 24-hour AI usage. Google’s greenhouse gas emissions rose by 48 per cent between 2019 and 2023, which is attributed in large part to a huge increase in data centre energy consumption.
Earlier this year, the World Economic Forum shared that the power that is required to sustain AI’s rise is doubling roughly every 100 days. To put that in context, it is estimated that, in 2028, AI will use more power than the entire country of Iceland used in 2021. AI is not a sustainable industry, and we should talk about whether it is possible for it to be so, and whether its growth is, therefore, worth the climate cost.
I will move on to disinformation. I am one of many frustrated Google loyalists who currently cannot find things as easily, now that AI has been heavily incorporated into search results. I look at some of the suggestions that come up and are pushed to the top of the page, and I cannot wrap my head around the extreme nonsense that I am reading. I am not sure that if I were nine years old and doing a school project, I would know that what I am reading, from a trusted source, might not be true. It is important that people understand the limitations of AI-generated information, and understand that it is not fact. It needs to be questioned and scrutinised, and critical thinking must be applied.
There are currently wars being fought around the world, and what is happening on the ground is not the only conflict. Money is being poured into bots, algorithms and AI-generated content to convince people that one side is committing crimes when it is not, or to present convincing propaganda. It is becoming harder and harder for even experts to identify AI-generated images, videos, articles or even people.
Generated content comes with other dangers, too. We have all heard stories of the use of AI in sexual harassment. Taylor Swift, who is probably one of the most powerful women in the world right now, has been a victim of that. AI-generated pornographic images of her were circulated on social media at an incredible speed, and were about as unavoidable as her discography was when she played Murrayfield. That form of sexual violence is now experienced by women and girls around the world in the workplace, on the internet and at school, including many who do not have access to the lawyers and the power that is needed to erase such images from social media.
Before this debate, Zero Tolerance reached out to me and shared some research that shows that non-consensual intimate images constitute 96 per cent of all deepfakes that are found online, and that 99.9 per cent of those images depict women. It is popular: one major deepfake sexual abuse website receives 14 million monthly visits. That is terrifying.
Addressing that requires us to address all forms of violence against women and girls, but it can also be mitigated through requiring and facilitating the removal of such content and regulating how algorithms are allowed to promote it.
Privacy fears extend beyond non-consensual imagery. The European Union is leading the way on AI regulation, and the limits on AI use in its artificial intelligence act include banning what it calls “unacceptable risk” use, such as the cognitive behavioural modification of people or vulnerable groups, biometric identification systems such as facial recognition and Government-run social scoring. One example of the latter involves a Chinese city that uses a system that can see residents’ social scores go down for acts such as spreading harmful information on forums. If we do not believe that “Black Mirror”-esque systems that see people punished for non-criminal activity should be allowed, now is the time for us to say so to AI developers and Governments the world over.
AI is already being used in healthcare and other industries. What rights should people have over the way that their data is used? What rights do content creators, artists and musicians have to protect their material from copy and re-creation by a non-entity?
Finally, I point out—as my motion does—that it is not just naysayers who are calling for regulation. AI creators are calling for the same: those are people who truly understand how fast moving the technology is.
A recent Massachusetts Institute of Technology study showed that AI models were beginning to exhibit deception, acting differently when under observation by those who were performing tests to ensure that they were not acting beyond expectations. It is already a difficult technology to control and monitor.
Every other week, there are stories of a misguided legislator somewhere in the world trying to limit social media and mobile phone use for young people. That horse has bolted; AI has not—yet—but we are very, very close.
In terms of what we in the chamber can imagine, AI’s capabilities may as well be limitless in theory. It is on us, on the United Kingdom Parliament and on other Parliaments around the world to define its limits in practice.