කෘත්‍රිම බුද්ධිය

 කෘත්‍රිම බුද්ධිය

 

මේ නිව් සයන්ටිස්ට් සඟරාවේ දැනුමේ සීමා පිළිබඳ  වූ ලිපි පහෙන් අවසාන ලිපියයි. ඒ කෘත්‍රිම බුද්ධිය පිළිබඳ ව. කෘත්‍රිම බුද්ධිය පරිගණක ඇසුරෙන් සිදුවන්නක්. ලංකාවේ නම් කොහොමටත් ඉන්නේ කෘත්‍රිම බුද්ධිමතුන්. ඒ කොහොම වෙතත් මෙහි කතා කරන්නේ දැනුම අභිසංස්කරණය ගැන නො වෙයි.  දැනුමේ සීමා වෙනස් කිරීම ගැන නො වෙයි.

 

How AI is shifting the limits of knowledge imposed by complexity

 

From weather to the structure of proteins, some things are predictable in theory, but too complex to figure out in practice. But the rise of artificial intelligence is changing that fast

PHYSICS 10 January 2023

By Anna Demming

 

Storms can still surprise forecasters

 

NASA

Everyone knows it is impossible to predict the future, but not a lot of people pause to wonder why. Even putting aside the issue of free choice, it isn’t straightforward. After all, Isaac Newton’s laws of motion can be used to calculate what any object will do if we know its starting trajectory and the forces acting on it. French thinker Pierre-Simon Laplace once imagined a being armed with these laws and a lot of information, writing that “for such an intellect, nothing would be uncertain and the future, just like the past, would be present before its eyes”.

 

The reason the world still unfolds in a cascade of the unexpected is that there is a gulf between what equations can predict in theory and what it is possible to calculate in practice. The limits of our technology, the speed of our computers and the incredible complexity of nature all mean that some things are practically impossible to know.

of knowledge, in which we explore:

 

How can we understand quantum reality if it is impossible to measure?

Why maths, our best tool to describe the universe, may be fallible

Why some aspects of physical reality must be experienced to be known

Logic underpins knowledge – but what if logic itself is flawed?

One problem is that the things we want to study are sometimes composed of many objects that mutually affect each other. So while we can predict the path of a flying football just fine, we can’t do the same with particles because there are usually lots of them flying around, all exerting forces on each other. It is beyond our current computing abilities to simulate all those interactions at once – with any more than about 10 particles, we don’t stand a chance.

Complexity is a problem in many fields of science, not least medicine. Take proteins, the long strings of amino acid molecules that fold themselves up into intricate shapes inside our bodies to do all manner of jobs, from powering our metabolism to fending off bacteria. We know what forces and considerations dictate the functional, folded shape each protein will adopt. But there are so many atoms interacting with each other that we can’t compute this perfectly. This is a frustrating limit to our knowledge, because knowing the precise structure of a protein can help us design new drugs.

 

Chaos theory

There is an even more fundamental issue at play. It turns out that the behaviour of some systems are sensitive to even the tiniest difference in starting conditions – they are subject to what we call chaos. The weather is a classic example. Small changes in air temperatures or moisture levels on one day can result in unpredictable storms the next. Chaos applies to apparently simpler scenarios too. The roughly 27-day orbit of the moon around our planet varies erratically by up to 15 hours from month to month due to the constantly shifting pull the moon feels from Earth and the sun.

 

Chaos and complexity certainly put limits on what we can know. But they are more malleable than the iron-clad boundaries dictated by physical laws themselves (see “How can we understand quantum reality if it is impossible to measure?”). For example, by measuring atmospheric conditions in the present more precisely and using faster computers, we can make better weather forecasts, up to a point.

 

While aspects of the world – from weather to financial markets to the patterns of disease spread – will always be subject to chaos, there are tricks we can pull to understand them better. One helpful strategy, says physicist Tim Palmer at the University of Oxford, is to run a large series of simulations of the system you are studying with tiny variations in your starting conditions. The sooner the outcomes of the scenarios begin to diverge, the more unpredictable the system is. These “ensemble simulations” are now a standard approach. This is why rain forecasts now often come with a percentage chance attached – helpful for giving us an appropriate level of confidence in leaving our umbrellas at home. “Enlightened ignorance is often characterised as knowing what you don’t know,” says philosopher James Ladyman at the University of Bristol, UK.

 

Read more: AlphaFold: Why DeepMind’s protein-folding AI is transformational

But it doesn’t necessarily have to be this way. A case in point is the way that artificial intelligence (AI) has revolutionised our ability to calculate the structures that proteins adopt. Deep-learning AIs don’t attempt to crack this by modelling the physics. Instead, they are trained on thousands of known protein structures and use this knowledge to predict new ones. In July 2022, AI firm DeepMind said that its AlphaFold algorithm had calculated the structures of 200 million proteins, nearly all of those known to science. Sometimes, the limits of what it is practically possible to fathom can change drastically in a blink.

 

Will we ever build a warp drive?

Most spacecraft use propulsion systems. But there might be another way to zip around: by warping the space around you. Think of it like swimming breaststroke, where you scoop up water in front of your body and push it backwards. Could we ever build an engine that does this in space?

 

Everything with mass warps space-time to some extent. The appeal of a warp drive is that, in theory, it could effectively move faster than light. To do that, however, it would have to be so massive that it would form a black hole. There is a workaround, proposed by physicist Miguel Alcubierre: use something with negative mass. This would warp space-time without creating a black hole. There are good reasons to think objects with negative mass can’t exist, but the laws of nature don’t rule them out. The upshot is that a useful warp drive will probably – but not definitively – never be possible.