Neural networks, inspired by our brain’s architecture, are scoring spectacular successes in gaming and pattern recognition. If you’re an iPhone X or advanced Android user, you’ll already find it pretty normal that the device can actually identify your face. This is just one example of ‘principal component analysis’ – using data to classify people, rather than human preconceptions. And thanks to machine learning, computers are getting smarter by the day. What’s ahead for Artificial General Intelligence?
Is AI set to out-smart us? What could the road from smart Artificial Intelligence to Artificial General Intelligence look like? Will it lead to wise decisions? In this 3rd article in our series exploring Wise Decision Making and AI, we take a look behind the scenes of the machines.
Putting the G in AI | 8 Points
True, generalized intelligence will be achieved when computers can do or learn anything that a human can. At the highest level, this will mean that computers aren’t just able to process the‘what,’ but understand the ‘why’ behind data — context, and cause and effect relationships. Even someday chieving consciousness. All of this will demand ethical and emotional intelligence.
1 - We underestimate ourselves
The human brain is amazingly general compared to any digital device yet developed. It processes bottom-up and top-down information, whereas AI (still) only works bottom-up, based on what it ‘sees’, working on specific, narrowly defined tasks. So, unlike humans, AI is not yet situationally aware, nuanced, or multi-dimensional.
2 - When can we expect AGI? Great minds do not think alike
Some eminent thinkers (and tech entrepreneurs) see true AGI as only a decade or two away. Others see it as science fiction — AI will more likely serve to amplify human intelligence, just as mechanical machines have amplified physical strength. See at a glance how they are positioned in the full article.
3 - AGI means moving from homo sapiens to homo deus
Reaching AGI has been described by the futurist Ray Kurzweil as ‘singularity’. At this point, humans should progress to the ‘trans-human’ stage: cyber-humans (electronically enhanced) or neuro-augmented (bio-genetically enhanced).
4 - The real risk with AGI is not malice, but unguided brilliance
A super-intelligent machine will be fantastically good at meeting its goals. Without a moral compass, AGI will be like a loose projectile on steroids.
5 - AI has to learn how to learn
AI applies supervised learning, and needs a lot of data to do so. Humans learn in a ‘self- supervised way’. We observe the world, and figure out how it works. We need less data, because we are able to understand facts and interpret them using metaphors. We can transfer our abilities from one brain path to another. And these are skills which AI will need if it is to progress to AGI.
6 - AI has to understand cause and effect
When they have access to large data sets, today’s neural networks or deep learning machines are super-powerful detectors of correlations and conditional robabilities. But they still can’t understand causality – the relationship between one thing and another. This ability to establish and understand causal models to grasp complex reality remains a human skill. Another one which AI will need to acquire, if it’s to reach AGI..
7 - It’ll need another giant step to get from causal thinking, to consciousness
Exactly how neurons interact, and which parts of the brain are responsible for human consciousness, remain unknowns. So how AGI will achieve consciousness is a big question. It’ll need to get to grips not just with causality but with counterfactuals (how a causal relationship would change, given the introduction of some other condition into the equation).
8 - The wise application of A(G)I will need a moral compass
Human abilities still lie beyond the outer rim of AI. As mentioned, they involve self- supervised learning, and understanding causality. Most fundamentally, they involve framing and answering ethical questions, feeling empathy and compassion. These abilities will be critical to ensuring that AI — and AGI - are applied wisely, in a way that is ethical, responsible and sustainable.
We’re therefore left with AI as a purely data-driven or statistical approach to the world — very powerful for prediction and perception tasks: pattern and voice recognition, image perception and control, such as driverless cars and robotics. Less so, in the knowledge space: reading contexts, motivations and causal thinking.
To all of this, we add a final, vital point: the human ability to frame and answer ethical questions, to feel empathy and compassion. These abilities still lie beyond the outer rim of AI. Yet they will be fundamental to ensuring that it is applied wisely, in a way that is ethical, responsible, and sustainable.
Today’s big data analytics and deep learning machines — all part of narrow AI - may lead to smarter decisions. But they cannot, at the current time, make wise(r) decisions. This remains a unique human ability, one that we need to apply more effectively if we are to sustain our physical habitat (our planet) and our socio-economic habitat (our society). Humans will never beat a computer in speed and data processing efficiency. But when it comes to creativity, intuition and therefore innovation, we are far superior.
Having the wisdom to apply our innovative power — in collaboration with AI-driven machines - doing so in a way that is ethically and environmentally sound - would be an incredible step forwards on this fascinating road.
Read the full article here.