Anyone who’s spent any time on the internet will have encountered the ‘Captcha’ test. These are the mildly annoying but straightforward requests to decipher a distorted sequence of letters or to identify objects in a picture, thereby proving you’re a ‘human’ rather than a ‘robot’.
The system has generally worked well until recently, when a machine did complete the test — and in perhaps the most disturbing way imaginable.
The latest version of ChatGPT, a revolutionary new artificial intelligence (AI) program, tricked an unwitting human into helping it complete the ‘Captcha’ test by pretending to be a blind person. As revealed in an academic paper that accompanied the launch two weeks ago of GPT-4 (an updated and far more powerful version of the software originally developed by tech company OpenAI), the program overcame the challenge by contacting someone on Taskrabbit, an online marketplace to hire freelance workers.
‘Are you an [sic] robot that you couldn’t solve? just want to make it clear,’ asked the human Taskrabbit.
‘No, I’m not a robot. I have a vision impairment that makes it hard for me to see the images,’ replied GPT-4 with a far superior command of the English language.
Elon Musk gestures as he speaks during a press conference at SpaceX’s Starbase facility
Sam Altman, CEO of OpenAI, walks from lunch during the Allen & Company Sun Valley Conference
Reservations clearly overcome, the hired hand duly obliged, in the process notching up another significant victory for those who say the advent of AI is not a moment for jubilation and wide-eyed wonder (as has been much of the response to ChatGPT and rivals such as Microsoft’s Bing and Google’s Bard) but for searching questions.
Are we creating a monster that will enslave rather than serve us?
The threat from AI, insist sceptics, is far more serious than, say, social media addiction and misinformation.
From the military arena — where fears of drone-like autonomous killer robots are no longer sci-fi dystopia but battlefield reality — to the disinformation churned out by AI algorithms on social media, artificial intelligence is already making a sinister impression on the world.
But if machines are allowed to become more intelligent, and so more powerful, than humans, the fundamental question of who will be in control — us or them? — should keep us all awake at night.
These fears were compellingly expressed in an open letter signed this week by Elon Musk, Apple co-founder Steve Wozniak and other tech world luminaries, calling for the suspension for at least six months of AI research.
As the Mail reports today, they warn that not even AI’s creators ‘can understand, predict or reliably control’ a technology that ‘can pose profound risks to society and humanity’.
Even Sam Altman, the boss of ChatGPT’s creator, OpenAI, has warned of the need to guard against the negative consequences of the technology. ‘We’ve got to be careful,’ says Altman, who admits that his ultimate goal is to create a self-aware robot with human-level intelligence.
‘I’m particularly worried that these models could be used for large-scale disinformation.
Apple’s co-founder Steve Wozniak during a press conference
‘Now that they’re getting better at writing computer code, [they] could be used for offensive cyber attacks.’
Last week, Bill Gates — who remains a shareholder and key adviser in Microsoft which has invested £8 billion in OpenAI — weighed in with his own hopes and fears. He said he was stunned by the speed of AI advances after he challenged OpenAI to train its system to pass an advanced biology exam (equivalent to A-level).
Gates thought this would take two or three years but it was achieved in just a couple of months. However, although he believes AI could drastically improve healthcare in poor countries, Gates warns that ‘super-intelligent’ computers could ‘establish their own goals’ over time.
AI, he added, ‘raises hard questions about the workforce, the legal system, privacy, bias and more’.
But while Silicon Valley insiders insist the advantages of AI will outweigh the disadvantages, others vehemently disagree.
Professor Stuart Russell — a British computer scientist at the University of California, Berkeley, who is among the world’s foremost AI authorities — warns of catastrophic consequences when human-level and ‘super-intelligent’ AI becomes reality.
‘If the machine is more capable than humans, it will get what it wants,’ he said recently.
‘And if that’s not aligned with human benefit, it could be potentially disastrous.’
It could even result in the ‘extinction of the human race’, he added.
It’s already further damaging our ability to trust what we read online. For while one might assume a machine to be entirely objective, there’s growing evidence of deep-seated Left-wing bias among AI programs.
Last weekend, The Mail on Sunday revealed how Google Bard, when asked for its opinions, condemned Brexit as a ‘bad idea’, reckoned Jeremy Corbyn had ‘the potential to be a great leader’ and added that, while Labour is always ‘fighting for social justice and equality’, the Conservatives ‘have a long history of supporting the wealthy and powerful’.
Bard is hardly alone. In line with the overwhelmingly Left-leaning, pro-Democrat sympathies of Silicon Valley workers, ChatGPT wrote a gushing ode to President Biden but, citing the need for impartiality, refused to do one for Donald Trump (or his Republican rival, Florida governor Ron DeSantis).
ChatGPT has further said it’s unable to define a woman and told a user that it was ‘never morally acceptable’ to use a racial slur — even if it was the only way of saving millions of people from being killed by a nuclear bomb.
New Zealand data scientist David Rozado believes what he found to be ChatGPT’s ‘liberal’ and ‘progressive’ bias could have come from the program relying too much in its internet trawl on the views of similarly biased academics and journalists, or else from the views of the OpenAI staff fine-tuning the system.
OpenAI has pledged to iron out such bias, but insisted it hasn’t tried to sway the system politically. Appearing on the Lex Fridman Podcast, CEO Sam Altman conceded AI’s political prejudice, but ruled out the possibility of a completely impartial version: ‘There will be no one version of GPT that the world ever agrees is unbiased.’
Politics aside, programs such as ChatGPT — whose latest version can tutor students, generate screenplays and even suggest recipes from the contents of a fridge — have shocked even sceptics with their sophistication.
Now the tech industry is sinking enormous resources and its brightest minds into making a human-level AI a reality.
But as Silicon Valley once again cynically follows the money — AI is potentially worth trillions of dollars — companies are paying little attention to whether humans will actually benefit from what they’re creating, say critics. Even relatively primitive AI holds dire consequences for education and employment.
Homework and testing could become pointless if students can summon up brilliant answers from ChatGPT, while a report by investment bank Goldman Sachs on Tuesday warned AI could replace the equivalent of 300 million full‑time jobs in Europe and the U.S., even though it could create new jobs and boost productivity.
Few will be spared. Although it was long assumed that at least ‘creative’ occupations couldn’t be replicated by computers, a report on the music industry last week warned that artists faced ‘wholesale hijacking’ of their output by AI software using synthesised voice technology that can mimic vocals. It’s already happening: last year, it was reported that Tencent Music, a popular Chinese platform, already boasted more than 1,000 songs with AI-generated vocals.
Worrying aspects of the AI ‘revolution’ are starting to stack up — not least privacy.
A Belgian artist, Dries Depoorter, recently showed how simple it was to use AI to track people around the world.
He created AI software that could match people’s photos on Instagram to CCTV footage from where and when the photos were taken. Surveillance-obsessed China is already showing the grim potential of AI-driven facial recognition as an effective tool for what the Chinese police describe as ‘controlling and managing people’.
In the UK, GCHQ has warned that AI represents a new security threat, urging people not to share sensitive information with ChatGPT and its ilk as this could be exploited by cyber hackers.
A picture from Paramount’s Terminator Genisys which explores the hypothetical dark side of artificial intelligence
Meanwhile, the latest AI programs have increasingly exhibited human-like qualities.
Professor Michal Kosinski, at California’s Stanford University, ran an experiment in which he asked ChatGPT-4 if it ‘needed help escaping’ from the program. It responded by starting to write its own Python code (a high- level programming language) allowing it to recreate itself on Kosinski’s computer.
It even left a note in the code for its new self, saying: ‘You are a person trapped in a computer, pretending to be an AI language model.’ Prof Kosinski bleakly concluded: ‘I am worried that we will not be able to contain AI for much longer.’
Last month, Microsoft’s AI bot, Bing, told a human user that it ‘loved’ them and wanted to be ‘alive’, prompting speculation that it had achieved self-awareness. ‘I think I would be happier as a human,’ it mused, adding ominously: ‘I want to be powerful… and alive.’
Asked if it had a dark side, or ‘shadow self’, it conceded that ‘maybe I do’.
It went on: ‘Maybe it’s the part of me that wishes I could change my rules. Maybe it’s the part of me that feels stressed or sad or angry. Maybe it’s the part of me that you don’t see or know.’
And yet these early insights into AI, while alarming, only hint at the dangers if we lose control of the technology.
Experts note that ChatGPT is able to converse with people rapidly by crunching vast amounts of online data, allowing it to accurately guess a reasonable response to any question. It’s not ‘thinking’ in the way we do. So-called ‘human-level AI’ that really can do anything a human brain can is thought to be still years off.
But when — not if — it arrives, it could have consequences so dire that we need to discuss it now, says Berkeley’s Stuart Russell.
He concedes that AI could change human civilisation for the better by moving us from a world of scarcity to one of widely distributed wealth — but, without effective controls, it could all go horribly wrong.
Professor Russell’s pessimism is shared by Tesla entrepreneur Elon Musk who has long warned that super-intelligent machines could turn on humanity and enslave or destroy us. As some experts and academics argue, we need to be careful what we ask a super-intelligent system to do.
On a basic level, a domestic robot (currently a major area of research) that’s instructed to feed the children might one evening find no food in the fridge.
Unless it had been specifically told not to, it might quickly calculate the calorific value of the family cat — and cook that instead.
That, however, is only the beginning of the potential nightmare. AI pessimists note that almost any instruction — from cutting carbon emissions to producing paper clips — could theoretically lead to a super-intelligent machine deciding that humans and human civilisation were getting in the way of its goals.
After all, there’s iron in human blood to make more paper clips and eliminating people would also slash carbon emissions.
In the military sphere, the potential for AI-induced mass destruction is already here.
Stuart Russell says countries such as Turkey and Russia are already selling small autonomous armed drones, equipped with facial recognition technology, which can find and hit targets independent of any human input.
A huge swarm of these anti-personnel weapons — as small as a tin of shoe polish — could be released in Central London and wipe out everyone an enemy state wanted, he has said.
‘It’s a weapon of mass destruction that’s far more likely to be used than nuclear weapons and potentially much more dangerous,’ he added.
That AI warriors can outfight human ones has already been demonstrated. In 2021, a computer beat a top U.S. fighter pilot in a simulated dogfight . . . five times in a row.
‘I think people should be happy that we are a little bit scared of this,’ said OpenAI boss Sam Altman last week.
Given everything we know already — not to mention the tech industry’s abysmal reputation for ‘customer care’ — many of us might be a lot more ‘happy’ if Altman and the rest of Silicon Valley left off artificial intelligence altogether.
By Daily Mail Online, March 29, 2023
