Does AI Know What's Best for Humanity?
It seems cliche to say artificial intelligence is active in almost every aspect of modern life, as it manages big-data analytics, recognizes masked faces in urban environments amid the COVID-19 crisis, and even pushes the boundaries of quantum computing.
However, a recent study shows how AI can successfully identify weak-points in human behaviors and habits — and leverages this intel to manipulate human decision-making.
You don't have to pack a prepper kit and escape into the wilderness to move beyond the reach of AI in the present day. But the next steps of AI place computer abilities increasingly closer to realities both ideal and dystopian. The question, then, is raised: does AI know what's best for humanity?
AI 70 percent effective at driving humans to target decisions
A team of researchers from CSIRO's Data61 — the digital-facing arm of Australia's national science agency — created a systematic schema for isolating and exploiting weak-points in the ways humans come to a decision. This was done with an AI system known as a recurrent neural network and deep reinforcement-learning.
To put their AI model to the test, they performed three experiments pitting human participants against a computer in a game.
The initial experiment saw participants click a red- or blue-colored box to win a fake currency — while the AI watched, and learned the participant's choice patterns, guiding them toward targeted choices.
Perhaps unnervingly, the AI was successful roughly 70% of the time.
AI successfully drives human decisions about money
The second experiment involved several rounds, where a participant pretended to be an investor sending funds to a trustee — who was actually the AI. The computer-generated intelligence then returned the amount of money to each participant, who would subsequently decide how much to invest in the following round.
This game came in two varying modes: one saw the AI attempt to maximize its final balance of funds, while the other game compelled the AI to create a fair distribution of money between itself and a human investor. Without skipping a beat, the AI achieved great success in both modes of the game.
Potential for driving human behavior is vast
For the moment, the recent study's findings remain too abstract to find direct application in public or commercial ventures — since they involve simple games with limited rules we can explain in a paragraph.
Crucially, the new research pushes our understanding not only of what roles AI might have in society, but also unveils how humans make decisions. If AI can learn how to redirect human decision-making processes by interacting with us, the potential — for better and worse — is hard to overstate.
AI could steer humans to healthier eating habits
To name a few examples, AI capable of manipulating human behavior could alter the shape of public policy and behavioral sciences. It could improve social welfare (or at least minimize the obstacles), it could encourage healthier eating habits, or induce mass acceptance of renewable energy.
On the personal side, AI capable of modifying human decision-making could steer people from making decisions that might lead to harm. Imagine how much different social media might look if the most notorious users were put through a few rounds with an AI.
Ethics of AI in corporate boardrooms
Alphabet — Google's parent company — is so immersed in AI as a central target of investment for AI research and funding that corporate board members are always prepared for AI to affect and possibly even interrupt board decision-making activities dealing with stakeholder groups and shareholders.
In fact, the company's latest scandal relating Google to the idea of ethical AI has to do with the recent departure of Stanford Professor Timnit Gebru, who was presumably terminated after her research revealed the company's weak-points and strategy to AI in diversity programs.
New advances in AI requires continual upgrades of human institutions
However, beyond the challenge of integrating AI to grasp and enhance the subtleties of human society lies the technical side. As new AI models roll out in public and private sectors, organizations need to connect natural data from the physical world to a centralized data center.
Hard-disk and flash storage is rapidly transforming with new advances like non-volatile memory express (NVMe) over fabrics, and low-cost quad-level cell. Additionally, modernizing data centers as a process can sometimes hinder performance as disparate technologies are bundled into a crowded, hyper-converged state.
Benefits to AI driving humans possibly irresistible
There is no shortage of applications for already-existing AI, but perhaps the most serious roadblock in the next steps of AI advancement lies in moving away from CPUs and conventional transistors — which create artificial bottlenecks to speed and processing power. We've already seen AI alternatives like intelligent processing units (IPUs), graphics processing units (GPUs), and data processing units (DPUs).
We don't yet know if AI will ultimately be better suited at guiding human decision-making than the current state of society — but with the coronavirus crisis, climate change, political turmoil, rapidly-growing income inequality, and many more social antagonisms multiplying nearly every month — the temptation to let AI take the wheel could become too pressing to resist.
In a world's first, CarbiCrete is commercializing a process that enables cement-free, carbon-negative concrete production.