The old religions of the world may be dying, but newer religions are rushing in to fill the vacuum. The religion of technology is now demanding a blind, unquestioning faith in its messiah of Artificial Intelligence. When it comes of age, this messiah will lead us to salvation, they say. It will ease our suffering, deliver us into trans-human heaven and make us eternal. It has many proselytising cardinals and the esteemed Ray Kurzweil is its uncontested pope. In fact, he has authored the book, ‘Spiritual machines ’, where he claims singularity (his take on enlightenment) will come soon, in 2045, when humans will merge with machines. A super-intelligent species will dominate the planet ,in what will be the new era of evolution. This AI led evolution will beat the slow and inefficient biological evolution.
Mr. Kurzweil predicts that human level AI will come by the year 2029, when the human brain will have been duplicated, but there are reasons to believe his predictions are unrealistic, for more reasons than one.
Before placing unbroken faith in these declarations, even if they come from such experts as him, the director of technology at Google, let us take a deeper look at the claims. So what if we are not experts? If blind belief in the word of religions qualifies for uneducated faith, blind belief in this could too! Educate yourselves, the scientists always patronise from rooftops. Why not explore the matter for ourselves?
Let’s do it then.
It is possible, you might say, convinced by the recent victory of Deep mind’s Alpha GO over the world GO champion. When a program has beaten a world champion at a game as complex and intuitive as GO, what is stopping it from evolving other facets of the human cognitive capabilities and intelligence? It definitely will, the AI team declares ,that once human level intelligence is reached, the next step to super intelligence will be quick.
The human brain requires 10^14 times less energy and is 10^3 times more efficient in pattern recognition and decision making, doing them with limited data as compared to current AI programs. However, neuromorphic programmers, who are coding the neural connections of the brain say that the accelerated version of AI will be 10^ 12 times more efficient than the human brain and will evolve 10^15 times faster than historic rates of evolution. These ideas sound promising, in principle, but let us examine the barriers to them. Considerable in variety and magnitude, they will have to be broken for and if the dream of singularity is to become a reality.
Different laws at different scales : Why are we assuming that the mind will be decoded at the classical level ?
It seems that the science of AI is walking into the old trap of focussing solely on the apparent, emergent phenomena to reveal all answers. Why are they assuming that our brain works completely on the classical level alone ? Why are they are assuming that neural synapses are just equal to circuits?
Take a look at the way multiple levels of reality EMERGE depending on scale, from a deeper , yet undiscovered layer. The hunt for this undiscovered fundamental layer is still in progress, with string theory and loop quantum gravity as its most promising contenders .
Notice that:
- All these layers have distinct , separate laws.
- The layers APPEAR real.
- They are only emergent /incidental/ epiphenomena of deeper fundamentals.
Could we be mistaking Emergent level phenomena for fundamental?
What is the problem in simulating the mind ,you might ask, even if it is an emergent phenomenon? If one can simulate a brain’s apparent functions, one can create a mind,no?
No.
Let’s take a relatable example, something that is easy to understand. So we have figured out certain properties of our space-time which we can manipulate and use. We decoded laws of motion and made cars and trains. We decoded gravity to build air planes and rockets. That worked because we did not need the whole picture, the comprehensive mechanics .
See , it worked locally, at that scale.
Now the problem began when sizes increased or decreased vastly. At very large scales , or at very small, Newton’s laws were not sufficient.They failed; something extra was needed. At very small and very large scales, the laws of motion and gravity are very different from what we normally understand. Imagine this, the mundane force of gravity has still not been decoded fully , and persists to be a thorn in the flesh for the modern physicist.
So at one level we know, but we don’t see the whole picture.
If physicists are asked to DUPLICATE or CREATE ‘space-time’ as a whole, they will not be able to do so with known laws. Far from Newton’s laws alone, QM and Relativity laws all of them together, are not sufficient. They need more , and they do not have it.
Do you see the analogy?
To copy and manipulate certain functions of the mind is possible now, and it might give the illusion that it is possible to manipulate and create the whole mind. However, that effort could fail, as we go deeper and smaller, as more complex simulations arise. Maybe mind simulation has not yet progressed to a level of complexity where such a situation may arise. Neuroscience and computing are both young fields, and maybe time will reveal the need for deeper fundamentals.
We might need something more then.
Will Approximate laws duplicate the mind accurately ?
I want to put greater stress on another important idea that was covered in a previous blog. The laws of reality we have currently are APPROXIMATIONS of the truth, they are just ESTIMATES of how the world functions. I would urge you to go through a previous blog to understand better, but lets look at another interesting example to elucidate.
Remember geometry from school , Euclid’s geometry?
Now Euclid’s geometry is the geometry of of plain surfaces. However,Einstein has long since proved that our space -time is a curved surface, not plain. Hence the angles of a triangle in true space do not really add up to 180º. That is an approximation. The Euclidean model fails on curved surfaces, for which new geometries have been successfully formulated.
You see, Euclid’s geometry works as a LOCAL approximation at the human scale, by simplifying curved space- time to plain, for ease. Unknowingly. In fact, it is not a logical necessity. Meaning, it is not some truth,out there, in nature.
Similarly, even Newton’s, QM or Relativity laws might not be logical necessities, once the more fundamental law comes.
Till then , they are workable approximations.
Then our understanding of neuroscience cannot be much more than an approximation , a rough estimate yet.
To claim then , by extrapolation of early success that human and superhuman level intelligence will come in the near future seems to be far stretch of fancy. Surely, there are no delusions about the fact that neuroscience and computing have a long way to maturity, even if their growth has been exponential over a very short span.
You might argue that it is unfair to predict the growth of neuromorphic science( programming the neural connections of the brain) on the basis of progress of physics , but it is only fair. Firstly because modern physics has been around for more than three centuries, having reached a high level of sophistication. Its progress and status- quo must be taken as indicators for the directions of any serious science.
Secondly, because other sciences seem to have met with a similar fate. There seems to be an alarming state of affairs, common to many fields of scientific endeavour right now. Progress has stalled, and not due to insufficient technological progress . In fact, the exponential progress of technology has not rocketed breakthroughs in many sciences, as was expected. They began with unbridled promise and euphoria on the back of a seemingly sound theories , but hit roadblocks from which they never truly recovered.
The theory of evolution is one such extremely poor approximation of reality, which seems right, but fails to provide workable mechanics. It cannot precisely explain anything, being only a roughly hewn theory with many gaping holes. Another is genetics, which has now morphed into epi-genetics , having found its own paradigm insufficient to explain human biology in totality. Genetics has been unsuccessful in explaining much , from physical characteristics to the causes of disease, inspite of having decoded a workable set of laws .
Maybe these mechanics are not enough because, first they represent emergent phenomena only, and second they are approximations. There could be undiscovered, more accurate mechanics, yet untapped by modern science.
In fact it seems quite likely there is a big problem of missing fundamentals.
The coming roadblock for mind simulation : Missing fundamentals
Pioneering engineers at IBM are building a ‘Brain in a Box’, by simulating the neural connections of the human brain. Other AI researchers contend that algorithms alone will be insufficient to produce understanding , and ’embodiment’ will be necessary to develop true AI. They believe robotics is the answer.
Brain in a box or robotics , the problem about to hit both these endeavours will be same. You see, even mapping biological structures does not give complete, comprehensive mechanics of their workings. This has been experienced in neuroscience and genetics already .
Neuroscientists have launched the ambitious ‘Human Connectome project’, where they intend to map all the 80 billion neurons and 100 trillion synapses of the brain, to be able to simulate them. Now, they have already successfully mapped the entire connectome of a basic ,1mm long worm called C Elegans decades ago. Note that this simple organism has 302 neurons and 7000 synapses only . Even though the mapping was successful, it added no game changing understanding about the functioning of the worm’s nervous system. It seems as if there are other missing fundamentals needed to explain the overall working of an organism , even as simple as the C. Elegans .
So then, we have two problems. First, it might take decades to map the human connectome, even with the exponential growth of computing. Mapping 80 billion neurons and 100 trillion synapses is a gargantuan task even by today’s computing standards. Second, say the mapping is completed, that might still not translate into a complete understanding of the brain’s functioning, far from being able to simulate or duplicate it. What’s the guarantee, when it has helped little in case of a uncomplicated organism ?
More Missing Fundamentals
When we talk of the Human Connectome project, we cannot avoid mentioning a similar big data undertaking – The Human Genome project. Scientists embarked on the endeavour of sequencing the 3 billion base pairs of human DNA, with everest high hopes of transforming human health by eradicating all diseases. The project began in 1993 and took 13 years to complete, yet it cannot be called a true success. Even though scientists succeeded in sequencing the genome, the science of genetics got much more murky; the new information created more confusion that clarity.They discovered that only 3% of DNA is involved in protein synthesis, i.e creation of the physical body, while the rest 97% has no known function, and hence called it ‘ Junk DNA’. The Junk DNA make no clear contribution to the human body, none we know of. Of the known genes, many do not contribute over 5% to a person’s physical attributes, like height and eye colour. No causal relationship between genes and diseases was ever discovered. They found that genes are only indicators of malfunction in the system, turning on or off , like switches ,under the influence of unknown factors. As a pre-eminent scientist says,“ In pointing at everything, genetics points at nothing”, and that there is a “Missing heritability problem”, which they cannot explain inspite of complete genomic sequencing .
It is obvious there are other fundamentals, which are missing.
The curious case of missing fundamentals continues
String theory and loop quantum gravity were the most serious contenders for a ‘Theory of Everything’ , but Physicists are not so sure now. Modern physics is now is a sort of cul-de-sac, unsure of the way ahead. In fact, detractors say that these theories are a mathematical mirage which has side-tracked a whole generation of physicists. There is certainly the problem of missing fundamentals here as well.
Sean Carrol, esteemed string theorist and science populiser, talks of the end of an era in physics,while Nima Arkani Hamed, one of the brightest contemporary minds says we will have to let established science go in order to move further.
You see, after the paradigm- shifting, earth- shattering discoveries by Einstein almost a century ago , no significantly radical idea has been able to disrupt and explain our reality more fully . Our current bunch of physicists are accomplished mathematicians and geometers , trying to fine tune and smoothen out already existing theories . This means, as Arkani Hamed says , the established theories may have to be discarded for good. He adds that the only solution is to approach the problems differently , from a different starting point .
We might have to look for different fundamentals to explain the same reality.
Fundamentals which are missing .