I want to suggest that you read the article by Michael Levin in the Sept. 3 issue of Noema Magazine on how our fear of AI’s potential is emblematic of humanity’s larger difficulty recognizing intelligence in unfamiliar guises. (One needs to be clear however, that AI of the GTP engines is not 'intelligence' in the broader sense of the term. They are large language models, LLMs.) Here are some clips from the later portions of his essay:
Why would natural evolution have an eternal monopoly on producing systems with preferences, goals and the intelligence to strive to meet them? How do you know that bodies whose construction includes engineered, rational input in addition to emergent physics, instead of exclusively random mutations (the mainstream picture of evolution), do not have what you mean by emotion, intelligence and an inner perspective?
Do cyborgs (at various percentage combinations of human brain and tech) have the magic that you have? Do single cells? Do we have a convincing, progress-generating story of why the chemical system of our cells, which is compatible with emotion, would be inaccessible to construction by other intelligences in comparison to the random meanderings of evolution?
We have somewhat of a handle on emergent complexity, but we have only begun to understand emergent cognition, which appears in places that are hard for us to accept. The inner life of partially (or wholly) engineered embodied action-perception agents is no more obvious (or limited) by looking at the algorithms that its engineers wrote than is our inner life derivable from the laws of chemistry that reductionists see when they zoom into our cells. The algorithmic picture of a “machine” is no more the whole story of engineered constructs, even simple ones, than are the laws of chemistry the whole story of human minds.
Figuring out how to relate to minds of unconventional origin — not just AI and robotics but also cells, organs, hybrots, cyborgs and many others — is an existential-level task for humanity as it matures.
Our current educational materials give people the false idea that they understand the limits of what different types of matter can do. The protagonist in the “Ex Machina” movie cuts himself to determine whether he is also a robotic being. Why does this matter so much to him? Because, like many people, if he were to find cogs and gears underneath his skin, he would suddenly feel lesser than, rather than considering the possibility that he embodied a leap-forward for non-organic matter. He trusts the conventional story of what intelligently arranged cogs and gears cannot do (but randomly mutated, selected protein hardware can) so much that he’s willing to give up his personal experience as a real, majestic being with consciousness and agency in the world.
The correct conclusion from such a discovery — “Huh, cool, I guess cogs and gears can form true minds!” — is inaccessible to many because the reductive story of inorganic matter is so ingrained. People often assume that though they cannot articulate it, someone knows why consciousness inhabits brains and is nowhere else. Cognitive science must be more careful and honest when exporting to society a story of where the gaps in knowledge lie and which assumptions about the substrate and origin of minds are up for revision.
It’s terrifying to consider how people will free themselves, mentally and physically, once we really let go of the pre-scientific notion that any benevolent intelligence planned for us to live in the miserable state of embodiment many on Earth face today. Expanding our scientific wisdom and our moral compassion will give everyone the tools to have the embodiment they want.
The people of that phase of human development will be hard to control. Is that the scariest part? Or is it the fact that they will challenge all of us to raise our game, to go beyond coasting on our defaults, by showing us what is possible? One can hide all these fears under macho facades of protecting real, honest-to-goodness humans and their relationships, but it’s transparent and it won’t hold.
Everything — not just technology, but also ethics — will change. Thus, my challenges to all of us are these. State your positive vision of the future — not just the ubiquitous lists of the fearful things you don’t want but specify what you do want. In 100 years, is humanity still burdened by disease, infirmity, the tyranny of deoxyribonucleic acid, and behavioral firmware developed for life on the savannah? What will a mature species’ mental frameworks look like?
“Other, unconventional minds are scary, if you are not sure of your own — its reality, its quality and its ability to offer value in ways that don’t depend on limiting others.”
Clarify your beliefs: Make explicit the reasons for your certainties about what different architectures can and cannot do; include cyborgs and aliens in the classifications that drive your ethics. I especially call upon anyone who is writing, reviewing or commenting on work in this field to be explicit about your stance on the cognitive status of the chemical system we call a paramecium, the ethical position of life-machine hybrids such as cyborgs, the specific magic thing that makes up “life” (if there is any), and the scientific and ethical utility of the crisp categories you wish to preserve.
Take your organicist ideas more seriously and find out how they enrich the world beyond the superficial, contingent limits of the products of random evolution. If you really think there is something in living beings that goes beyond all machine metaphors, commit to this idea and investigate what other systems, beyond our evo-neuro-chauvinist assumptions, might also have this emergent cognition.
Consider that the beautiful, ineffable qualities of inner perspective and goal-directedness may manifest far more broadly than is easily recognized. Question your unwarranted confidence in what “mere matter” can do, and entertain the humility of emergent cognition, not just emergent complexity. Recognize the kinship we have with other minds and the fact that all learning requires your past self to be modified and replaced by an improved, new version. Rejoice in the opportunity for growth and change and take responsibility for guiding the nature of that change.
Go further — past the facile stories of what could go wrong in the future and paint the future you do want to work toward. Transcend scarcity and redistribution of limited resources, and help grow the pot. It’s not just for you — it’s for your children and for future generations, who deserve the right to live in a world unbounded by ancient, pre-scientific ideas and their stranglehold on our imaginations, abilities, and ethics.
I don't even know where to start. To suggest that what is basically your phone keyboard's predictive text on steroids (with massive amounts of energy expended) is "intelligence" kinda boggles my mind. LLMs (and it's image generating cousins) are not intelligence. They do not *understand* the data they are using. They are simply using it to make predictions. And, to add to it, our human brains are wired to make meaning of things, so we then take the product of these predictive models, and see them as having meaning.
ReplyDeleteI don't actually know whether human beings are capable of making machines that *understand* and *make meaning*. Perhaps. Or perhaps that is a quality of wetware we may never manage to imitate. But until and unless our machines *understand* it is marketing hype only to call it intelligence.