Friday 27 February 2015

What future?

My good friend and colleague Alison Feist recently tweeted a link to a Guardian article.

Alison Feist @AlisonFeist1 · Feb 24
Could computers ever replace teachers? http://gu.com/p/46vzp/stw

This brought me back to a train of thought that briefly surfaced while writing my last EdD paper. Machines have already augmented learning and replaced some aspects of what teachers used to do. In some respects the replacing of teachers with machines is a relatively minor concern. The evolution of intelligent technologies is well underway, in the not too long term this offers potential for major transformation of our ability to augment natural learning, to repair humans, to understand the universe and all that is in it and to reach out and inhabit other planets. Wonderful and exciting developments but I do feel there are potentially high risks as far as human survival goes.

I was a special guest at a TELMAP meeting in Brussels in 2013 in which I facilitated several group discussions on future evolution of technology and learning, below is a summary of my group's thoughts.



We speculated about the possibility of machine / man symbiosis and largely assumed benevolence for the purpose of the exercise. To pursue consideration of the potential for a dystopian future would have been well off the appreciative vector the meeting was examining. The above table was a best shot in the time available to reflect core issues from the group discussion. Like others in my group I felt there was inadequate time to do the topic justice or to articulate fully the findings of discussion or the unanswered questions that it provoked. Two years on and it is clear that "Increasingly intelligent and ambient technologies" are appearing and that much effort is going into this evolution. A snapshot from today's media shows a few of these.

During our future gazing we briefly discussed the potential for chips in the brain able to read thoughts and transmit them to machines or other humans and felt that could lead to all sorts of interesting and worrying potentials. This is an area that is already progressing rapidly. Controlling a drone via brain remote control of commercial / military aircraft is almost here "Technology is evolving, regulations are evolving. [Unmanned jets are] obviously going to happen. The question is not if, it's when." A bionic eye linked to a chip implanted in the brain Human trials still a decade away but another important step towards synthesis of biology and machine. A machine teaching itself how to play games What caught my attention here was the link to intuition and processing / synthesis of large amounts of data: "Scientists are developing computer programs that - like the human brain - can be exposed to large amounts of data, such as images or sounds, and then intuitively extract useful information or patterns." The paper published in Nature is well worth a read; the 'deep Q network agent' is designed to combine reinforcement learning with advanced deep neural networks and seems to operate in a very similar way to biological systems building up abstract representations of data in order to learn concepts. The goal of the agent is to select actions that maximise reward. Although still in early days there is a lot of interest in imbuing robots with empathy, incorporating AI technology developed for Watson into robots is on the horizon.

Basic biological computers using slime molds and organic nano scale computing and the potential for brain to brain networks are being explored. Whether this is done by using existing animal brains or attempts to grow 'human' brain material, as we already do with other organs, it is a strange and slightly uncomfortable line of development.

I don't adhere to any religious concepts so for me the human brain is an autonomous processing machine and reward based intuition is exactly what these guys are talking about in hard technology machines. The emotional states associated with intuitive moments seem to me nothing more than a reward tagging mechanism. Success from an intuitive intervention feels good, feels clever, it feels like being 'tuned in'. 'It felt good so it is good and that is something I need to do again' is a key affective driver of our form of sentience and in my opinion is one shared by many other life forms. Empathy, another human characteristic, seems to be viable via machine. I don't see the need for a soul or other supernatural force to guide such insights. A powerful information processor with access to masses of past data that is compared with an in the moment experience could combine induction, deduction and abduction to arrive at the most likely solution/explanation/successful action. Although intuition and empathy often lead to a feeling of certainty or a 'knowing that I am right', a well developed critical thinker will also embrace uncertainty and be aware of caveats in relation to to intuitive or empathetic interpretations. The potential for this to be incorporated in everyday machines will be upon us in the very near future. Whether this is leading towards self aware or sentient machine 'life' in the near future is difficult to predict. The debate as to the nature of self awareness is complex and not satisfactorily resolved in my opinion. I am both excited and scared at the prospect of exploring this in machines.

Stephen Hawking recently warned that in relation to processing power and machine capability "there are no fundamental limits to what can be achieved" his assertion that humans could rapidly be outstripped by machines and that this might not be a good thing has been a theme in many science fiction works, this is a genre that is rarely take seriously by the larger population but is peppered with many ideas that have since been realised and many that may well be soon upon us. There were many flaws and flights of fancy in the imaginings of Sci-Fi authors but there were also many well thought out scenarios and messages. A key theme in 2001 A Space Odyssey is autonomy of machines. The film was influenced strongly by Kubrick's direction but was based on themes Arthur C. Clarke had explored in several books. HAL (Heuristically programmed ALgorithmic Computer) is in control of a spaceship; he starts 'behaving badly'. The initial programming and the logic inherent in the computer placed it's 'self' as infallible and humans as error prone with their lack of consistent logic. HAL appears to test Frank Poole by making a deliberate mistake in their game of chess and Poole fails to notice. This game is based on an outstanding game played in the 1930s. Poole's response suggests to HAL that humans are inferior in detecting error and hence are a potential risk to the mission. HAL goes on to question mission commander Dave Bowman and realises that he is not aware of the true purpose of the mission. The success of the mission is a deep element of HAL's programming, it could be conjectured that HAL felt this to be his reason for existing. With the potential for humans making errors and their lack of information about the real goal of the mission (that HAL does 'know') the logical step is to eliminate the humans to avoid mission failure. Hawking's assertion implies machines controlling their own destiny is a dangerous future, this is a timely reminder of a potential that authors like Clarke were offering many decades ago. HAL discovers that Poole and Bowman intend deactivating him and takes action to avoid that outcome. With careful engineering of situations by HAL the plot leads to point where Poole is dead and Bowman is outside the spacecraft. HAL refuses to let him in stating that he can not let the disconnection happen. Kubrick interpreted the disconnection scene very well with HAL displaying some quasi human characteristics attempting to deceive and playing for sympathy by pretty much pleading temporary insanity then trying to instill confidence in his ability "I feel much better now".

HAL is a useful illustration of the potential dangers of the evolution towards increasingly intelligent and ambient technologies and the development of a nature like relationship. It is not many decades since computers with 4Kb RAM filled whole floors of buildings and required punch cards to input information. From the late 80s development headed towards supercomputers Deep Thought then Deep Blue and Watson. These are hugely powerful but again not the kind of tech that could be bought by the general public. It is only a few decades since bulky desktop machines with 32kb of RAM revolutionised offices and education, it is not much less than 2 decades since laptops heralded the start of usable portable computing at a price that made them viable for widespread public ownership. The first smart phone arrived mid 1990s but was very basic and barely usable away from a power source. By 2007 the first iPhone arrived and small portable technologies started to mature. Apple, Pebble and others have developed smart watches into useful tools at a reasonable price and the functionality will increase rapidly as miniaturisation improves and near field communication becomes standard. Playful and useful tech is built into clothes and Google Glass could be seen as a very important stepping stone despite the problems inherent in the first iteration and the obvious tricky social considerations. Recently mind controlled prosthetic limbs are becoming very much usable. The increase in power and reduction in size of tech has reduced the intrusion factor, the ability to talk to a smart phone and for it to talk back gives a more life like aspect and as wearables and the internet of things develop technology is becoming part of the ambiance that surrounds us. The development of intuition and empathy will bring more life like aspects to our relationship with machines in the near future. However there is a lot of future ahead of us and if Moore's law stands up for even another decade it is hard to imagine what machines will be capable of.

Quantum computers have been a difficult goal to achieve but predictions that we are within 5 or 10 years of success are becoming more convincing. Quantum states are complex and move on from being either a 1 or a 0 state by introducing the potential to be both 1 and 0 simultaneously. There are many problems superconductors tend to only work at very cold temperatures so need vast cooling systems but research is showing a potential for new materials to escape this need. Until recently maintaining qbits in stable states longer than microseconds was impossible due to their spin state being vulnerable to change by the influence of nearby particles. Recent progress by Morello and Dzurak in 2014 isolating qubits in pure silicon seems to have achieved stability for up to 30 seconds. Reductions in error rates are leading to claims in 2014 / 15 that there are now no barriers and Google have set out with a team of 'quantum engineers' with the aim of developing a quantum computer. Entanglement of numerous qbits further multiplies the processing power but is difficult to achieve. Entanglement is easier to maintain with photons and progress at MIT in the development of chips coated with multiple single photon detectors is a crucial step towards achieving viable quantum computers. Other developments at MIT include swarms of drones that communicate and behave like insects. Power and autonomy are developing fast.

By 'life-like' my TELMAP group were also acknowledging that there is potential for both benevolence and aggression, altruism and selfishness, the potential for machines to develop notions of 'us and them' that underlie intolerance, racism, war, genocide, eugenics, and all the terrible achievements of our species.

We already have robots capable of building furniture, of self organising into swarms, able to repair and self repair. In 2005 a proof of concept basic self replicating robot was built at Cornell. A decade later the University of Oslo are experimenting with self learning, self repairing and self reproduction "The arms of one of the robots is fitted with a printer. This produces a new robot, or a new part for the existing robot, which enables it to negotiate the stairs.

The ability to take in sensory information and to process and store that information in a connection rich habitat such as an organic brain or primitive nervous system seems a key requirement for sentient life as we know it, this would also seem a potential that machine processing and storage systems could achieve. We seem to be on the brink of huge breakthroughs in synthesis of biological and machine interfaces, however these might well be a stepping stone to a very different existence. Given the vast distances involved, to conquer the stars, to explore even our own tiny galaxy in our current biological form is not feasible unless we can achieve light speed travel. Far easier would be to set intelligent machines off on that journey be they mechanical or bio-mechanical.

Autonomous reproduction and development of 'improved versions' is already here albeit in fairly simple machines. Similar developments are going ahead with nanotechnology. The unforeseen outcomes of the logic of HAL's programming may have been fictional conjectures but the potential for autonomous machine intelligence to excel (from its perspective) while malfunctioning (from our perspective) is a clear and present danger and is not one we should blithely ignore. Setting them off to explore the far reaches of space we would lose any vestige of control, it would be difficult to predict what form or attitude they might have should they return.

Looking beyond human intelligence and machine AI is what Clarke and Kubrick seem to have been concerned with in the final scenes in 2001 with the 'evolutionary rebirth' of Bowman as the star child. The speculation here is a new frontier in evolution. Positing that sentient existence is not exclusive to biological entities and that both organic and machine capacity is limited by material physics then the next logical step is existence in a pure energy form. Perhaps such an entity could travel anywhere and like qubits exist in more than one state at any time, If time turns out to be a perceived rather than actual dimension perhaps existence everywhere and every time could be possible just not in a way that we in our present state can detect.

I kindof set out wanting to record a trawl of recent developments in computing to consider whether we were anywhere near computers replacing teachers. I never imagined I would rediscover HAL and get even more scared and excited about our future that I already was. It has been an interesting journey.

No comments:

Post a Comment

Note: only a member of this blog may post a comment.