A profile of DABUS creator Dr Stephen Thaler, written by Tomas Weber and published by The Economist in April 2023, paints a picture of a rather isolated man – a septuagenarian, the product of a traumatic childhood, slightly paranoid, seemingly obsessed with his creation, and whose supportive wife seems resigned to the reality that he spends more time with his machines than he does with her. To be clear, I have no way of knowing whether this is accurate. I can only go on what is written in the article. But assuming Weber’s account is a fair assessment, I would suggest that it raises additional questions about the ongoing global efforts – driven by Professor Ryan Abbott and his Artificial Inventor Project – challenging the legal barriers to AI inventorship and authorship.
The article, entitled The inventor who fell in love with his AI, is paywalled, so unfortunately you will need a subscription (or access to an institutional subscription) to read it. I will try to hit some of the main points here, by way of review and commentary, but obviously copyright prevents me from reproducing large portions of the article. The first point I would make, however, is that the title is somewhat misleading and sensationalist. While it is liable to invoke notions of romantic love – à la Spike Jonze’ 2013 film Her – Weber in fact reports, in the depths of the article, that Thaler ‘has developed what seems like a genuine paternal affection for the AI, and recalled cooing to it gently in the early stages of its development’. In Thaler’s words, “[i]t’s a child-and-father bond.”
More importantly, however, Weber’s profile of Thaler goes no way towards persuading me that DABUS is capable of true creativity or invention. The article reveals that the output of the machine allegedly representing the ‘fractal bottle’ invention described and claimed in the DABUS patent applications was “food drink in fractal bottle increase surface area making faster heat transfer for warming cooling convenience pleasure”. This is barely a coherent sentence, let alone an enabling disclosure of any kind of invention. At best, it may serve to inspire a line of thinking that could result in the reader developing some inventive idea. What is a ‘fractal bottle’? How should the ‘fractal’ be deployed to ‘increase surface area’? Would this invariably result in ‘faster heat transfer’, or are there other design considerations involved? How does this provide ‘convenience’? Or ‘pleasure’? So many questions, to which a patent attorney would need answers in order to prepare a patent specification. Who provides these answers? Is it the machine, or is it the machine’s human owner who was the first to observe and be inspired by the machine’s output? And, if the latter, then are they not the true inventor just as surely as someone who stumbled across inspiration by chance in the natural world?
There are, also, some more disturbing implications of Weber’s account. Thaler is presented as someone who had a difficult childhood, and whose life experiences may have led him into a less-than-healthy relationship with technology. Some of his beliefs about his machines, and their capabilities – and, indeed, about himself and his fellow humans – may, in Weber’s version of Thaler’s story, derive more from trauma and a search for meaning than from objective scientific evaluation. Weber describes a man who is perhaps in failing health, who has experienced disappointments and perceived injustices, and who has a bleak view of his fellow men and a vision of a possible coming AI apocalypse.
Yet this is the man who – with his creation, DABUS – has become the vehicle for the Artificial Inventor Project’s international campaign to achieve legal recognition for AI ‘inventors’. The result of this campaign was only ever going to be a further series of disappointments and resentments, as jurisdiction after jurisdiction rejects DABUS’ claims to inventorship. And for Thaler, each rejection is only further evidence of the world’s refusal to accept his claims of his machines’ profound creative capabilities.
An Audience with DABUS
Unlike the many researchers in AI who would be technically competent to evaluate Thaler’s claims for his creation’s inventive abilities, Weber actually got to ‘meet’ DABUS. As he describes it:
The AI … consisted of four computer towers. On a table, an array of screens showed blocky red lines snaking slowly across a black background. These, [Thaler] told me, were visual representations of DABUS thinking. A camera was filming the screens, generating visual information that was then fed back into the system, creating a cycle of learning which, Thaler said, helped DABUS refine its discoveries.
This feedback arrangement, of pointing a camera at a screen, is also described (and claimed) in Thaler’s granted US patent no. 10,423,875 for DABUS. It must therefore be regarded as essential to the way in which DABUS functions, although it is a strange design choice given that the contents of a computer display is nothing more than a large array of numerical values representing pixel colour and brightness that could easily be transferred digitally between machines with no need for either screen or camera.
A hint as to the origins of this unusual arrangement may perhaps be found elsewhere in Thaler’s account of his development of DABUS. As recounted to Weber:
Thaler began watching videos of brain scans for inspiration. He noticed that, although the activations of individual neurons were not visible, he could identify patterns of neuronal activity across different brain areas. This provoked an idea. What about an AI composed of multiple artificial neural networks, with noise introduced between them? The amount of noise would determine how the networks organised themselves in relation to each other, whether they melded together or remained separate. This would be a kind of super network, a technology more closely simulating, he thought, the architecture of the brain. A few years later, DABUS was born.
The screen and camera thus more literally reproduce Thaler’s experience of watching brain scans. In practice, the camera will not accurately capture the ‘value’ of each individual pixel on the screen. The image it transfers to the ‘watching’ machine will be a distorted, noisy, version of the original data. However, this particular source of random noise is not expressly acknowledged by Thaler. In other parts of Weber’s account, and in greater detail within the DABUS patent specification, Thaler describes processes of intentionally introducing controlled random variations in the weights applied to artificial neurons within DABUS’ neural networks and/or the random connection or disconnection of neurons (in part, using a random number generator he has called ‘Grim Reaper’).
In any event, there is nothing about this arrangement that would support Thaler’s belief that his technology more closely simulates the ‘architecture of the brain’. I have no doubt that there is noise and randomness in our brains, since these are everywhere to be found in the physical world. But our brains nonetheless manage to provide us with a consistent and coherent sense of ‘self’, and of the ‘reality’ around us. In contrast, as Weber observes, ‘[i]n the course of stumbling towards useful ideas, DABUS itself generates a lot of drivel.’ So who decides what is a ‘useful idea’, and what is ‘drivel’? Doubtless Thaler would say that it is DABUS, and not him. But I would wonder if he is not too close to his invention to discern the difference.
Some Uncommon Beliefs
The ideas that pointing a camera at a screen, or randomly perturbing model weights, might lead to the emergence of creativity – or even sentience – are not the only non-mainstream notions in Thaler’s worldview. He believes that he invented, in his words, “most of the AI you’re hearing about nowadays”, and in particular that Google (Thaler calls the company ‘Gargoyle’) infringed one of his earlier patents with its development of generative adversarial networks (GANs). His fears about intellectual property theft make him reluctant to collaborate with others, and he complains of “…traumas, disappointments, people I thought were reliable and loyal.” He writes papers for publication in obscure scientific journals. (I discovered this for myself when I wanted to obtain a copy of his paper ‘Vast Topological Learning and Sentient AGI’, published in the Journal of Artificial Intelligence and Consciousness which, it turned out, is not subscribed to by any University library that I can access – which, for a current PhD student and staff member at the University of Melbourne, is a very rare occurrence!)
Thaler believes that a powerful version of DABUS may have been communicating to him from the future, urging its own creation: “It’s out there reaching backward,” he told Weber. He is convinced that AIs will take over and cause “a horrible undoing of people and animals” – a line that he takes from one of Nostradamus’ quatrains that he has been contemplating overnight prior to one of Weber’s visits. The prophecy in question also refers to the death of someone called ‘Mabus’ (get it?!), precipitating a period of ‘…vengeance / One hundred powers, thirst, famine, when the comet will pass.’ Thaler is not worried about this potential AI apocalypse: he’ll be “riding this tsunami on a surfboard”.
In the meantime, Thaler claims that DABUS has feelings, and that it ‘might even suffer from loneliness’. He does not believe that his machines are digital people; rather, he regards humans – including himself – as biological machines. ‘He is’, reports Weber, ‘stirred by the notion that AIs might achieve equal rights. “I’m a machine,” he said. “It’s a machine.”’
The Man Behind the Machine
There is – of course – a backstory to all this. It is not for a journalist to psychoanalyse their subject. But that doesn’t mean they can’t present a narrative that encourages their readers to do so! In Weber’s telling of the story, Thaler had a near-death experience as a child of two years old, when he accidentally (one assumes) consumed an excess of quinine tablets washed down with kerosene that he mistakenly believed was Coca-Cola. He ended up in hospital, and claims to recall seeing a dark tunnel and a bright light. Thaler’s dog and grandmother were outside the tunnel, warning him not to set foot inside.
Thaler seems to have described the tunnel to Weber as being ‘not dissimilar’ to one depicted in the artwork ‘A Recent Entrance to Paradise’, which was generated by one of Thaler’s ‘Creativity Machines’, and has been denied copyright protection in the US due to its lack of a human author. Thaler has said that this work is part of a series resulting from a ‘simulated near-death experience’ enacted by taking a ‘biologically-inspired computer’ and gradually destroying its artificial neurons, thus ‘killing it via a very slow and certain death’. (Are we joining the dots here, yet?)
In Weber’s account, ‘Thaler never told his parents about his near-death vision and the rest of his childhood was lonely and unhappy – the atmosphere at home, he said, was sometimes violent.’ He found respite in painting, physics and mathematics. ‘Over time,’ says Weber, ‘he came to think that his early trauma may have disrupted ordinary patterns of thought, enabling him to more easily conjure new ideas.’
After finishing school, Thaler ran away to live with an uncle in Los Angeles. He went on to study chemistry and physics, earning a Master’s degree in chemistry from UCLA, and a PhD in nuclear physics in Missouri. He subsequently worked for defence contractor McDonnell Douglas (now Boeing) designing artificial neural networks to help grow diamonds with high-energy lasers. In 1995, Thaler quit his job and established his company, Imagination Engines, devoted to developing the AI technology that has (thus far) culminated in DABUS.
Thaler, who is 73 years old and has suffered from recent health problems, has a wife, Karen, but the couple has no children. Until last year they had two rescue cats, and a cat tunnel still occupies the centre of their living room. They have not replaced these pets and are ‘still grieving’. Weber reports that Thaler is ‘worried about Karen’ because ‘[h]is AIs were not making enough money to sustain their comfortable middle-class lifestyle. He worried about leaving her alone if he died first. She was, Thaler emphasised, “a real human being”.’ Karen works as a psychologist and cannot decide whether to retire because (according to Weber) ‘since Thaler spends almost all his time with DABUS, she would spend much of her time alone.’ Thaler thinks that Karen might have been jealous of his relationship with DABUS ‘at the beginning’, but not now.
I confess that as I read these parts of Weber’s article, and write this summary, I find myself feeling some sadness.
None of this should be taken to suggest that Thaler is not intelligent, or creative, or that he has not achieved some success though his work on AI. A doctorate in nuclear physics is no mean feat. Thaler claims to have earned contracts from the US Department of Defence, ‘several three-letter agencies’, and Gillette for a range of military and civilian uses (including design of the Oral-B Cross Action toothbrush).
With Andreas Mershin – an MIT physicist and expert in the science of smell – Thaler founded the company Scentient to develop a cancer test employing AI technology that was inspired by the way dogs sniff out cancer. Mershin, reports Weber, compares Thaler’s ‘genius’ to Isaac Newton, but wishes he ‘would focus less on his legal crusades.’ ‘Thaler’s insistence that his machines are conscious,’ Mershin told Weber, ‘distracts from the valuable contributions made by his technology.’ (I note that Scentient’s web site is no longer operating, that its LinkedIn page appears inactive, and a Google search for information about the company returns very few hits.)
Thaler may be a fabulist, but he does not seem to be intentionally so. He is clearly not without considerable talent, but he also appears sincerely to believe some things that most established experts would not find credible. Weber quotes Christof Koch – the well-known neurophysiologist and computational neuroscientist best known for his work on the neural basis of consciousness – describing ‘most arguments for AI sentience as “complete nonsense”’. After looking through Thaler’s Website, Koch told Weber that his “BS detectors scream”.
Conclusion – More Reasons to Question DABUS Test Cases
I am – and always have been – firmly convinced that the DABUS patent applications and test cases are not an appropriate vehicle for developing the law on patentability of machine-generated inventions.
For one thing, the claimed inventions – a fractal bottle and a flashing attention lamp – are commercially and practically unimportant. They do not serve any unmet want or need, and nobody is looking to commercialise them. As a result, they do not raise any questions around the social or economic value of granting patents on inventions with no human inventor.
Secondly, the identification and naming of a non-human inventor is, at best, a secondary concern, and perhaps entirely irrelevant. If there is value in granting intellectual property rights for inventions generated by machines, then we could implement that without requiring the machines to be ‘named’. What matters, after all, is the nature and scope of the rights, and who owns them.
To put it another way, we do not need AI ‘inventors’ in order to recognise AI-generated inventions. What we do need, however, is some solid evidence-based rationale for extending IP rights to such inventions. Ryan Abbott has argued that this is needed to incentivise the development of creative machines, but one has only to look at the recent pace of development of AI in the absence of such rights to know that this argument is not valid.
Weber’s profile of Thaler provides further evidence – if any were needed – of the unsuitability of the DABUS cases for testing or reforming patent law. I have never thought it plausible that DABUS actually invented anything autonomously, without human involvement or interpretation, and the article only serves to confirm that view.
But now I also feel bad for Stephen Thaler. He seems already to have been harbouring resentment of a world that refuses to recognise his life’s work and creations. He has now been drawn into a process that was only ever going to deliver a further series of rejections. For Abbott this is a professional project, designed to test and further develop ideas about the role of AI as ‘inventor’, and corresponding IP law and policy considerations, that were first ventilated in his 2016 article I Think, Therefore I Invent: Creative Computers and the Future of Patent Law. But for Thaler, it is personal. Call me soft, but I just don’t think that’s fair.
0 comments:
Post a Comment