Reckon AI is coming? So is super-intelligence.

Technology research group Nakano’s chief analyst, Andrew Sheehy, is part of a movement to increase the business community’s understanding of what artificial intelligence (AI) is and what positives can come out of utilising AI in business. In this opinion article, he takes it a step further: superintelligence. >>  

GIVEN that the field of machine intelligence, or artificial intelligence (AI) is still clouded by controversy, then it might seem a little premature to say that ‘superintelligence’ is inevitable.

After all, many people would argue that we still do not know what human intelligence or artificial intelligence really are.

The approach I’m going to use here is to bypass completely the conventional arguments about machine intelligence and try to think about the subject in an unconventional way.

The starting point is an assumption that machine intelligence exists. The strength of the machine intelligence could be extremely weak and it could also be very limited but at this stage all we need to do is to accept that advanced AI systems like IBM Watson, or AplhaGo do indeed demonstrate at least some level of genuine intelligence.

But the moment we assume that AI systems like Watson and AlphGo exhibit some level of intelligence then we should immediately ask where that intelligence comes from?

An AI system can be divided into two parts: hardware and software.

The hardware is easiest to understand. Even the most complex computer system is untimely just a collection of semiconductors, circuit boards, passive components, plastics and metal.

It would be very hard to argue that the computer hardware itself possesses any intrinsic intelligence:

We could say that the hardware is intelligent in a way because it is the result of collective human intelligence – which was needed to conceive the design and figure out how to convert the input raw materials into the finished computer. And so if we added up all of the intelligences needed to produce the finished computer then the hardware could be said to embody that intelligence.

The problem with this is that the computer hardware, by itself, does not actually do anything, and nor can it do anything. If it cannot do anything by itself then it cannot exhibit any behaviour which we might deem ‘intelligent’.

In essence, the hardware is like a person the moment after death: intelligence was present, but it no longer is.

Hence, computer hardware as we presently understand it cannot possess intelligence.

The very most we can say about the hardware is that it is the physical environment where machine intelligence can reveal its presence.

The other part of the computer is of course the software.  But first, what exactly is ‘software’?

If this seems like a silly question then consider the following definitions:

• Software is the part of the computer that is not physical.
• Software refers to the programs and data used by a computer.
• Software is encoded information and computer instructions.

But while all of these seem to be reasonable definitions, none clearly define what the software actually is.

Let’s think about this a bit more carefully.

Programmers spend their time creating software which looks like:

  function __construct($settings) {

    $this->host = $settings['host'];

    $this->db   = $settings['db'];

    $this->user = $settings['user'];

    $this->pass = $settings['pass'];

    $this->charset = $settings['charset'];

    $this->connect();

  }

  function connect() {

    try {

      // connect to database

      $this->dsn = "mysql:host=$this->host;dbname=$this->db;charset=$this->charset";

      $opt = [

          PDO::ATTR_ERRMODE  => PDO::ERRMODE_EXCEPTION,

          PDO::ATTR_DEFAULT_FETCH_MODE => PDO::FETCH_ASSOC,

          PDO::ATTR_EMULATE_PREPARES   => false,

      ];

      $this->conn = new PDO($this->dsn, $this->user, $this->pass, $opt);

      $this->status = true;

    } catch (Exception $e) {

      print "Cannot connect to database, error was: ".$e->getMessage();

    }

  }

 

=========================================

But if we think about this carefully then it becomes clear that these are simply two examples of the many ways in which the software can be represented in our physical reality.

Now I know what you’re thinking, which is that the actual software resides in the memory of the computer where it ultimately exists as a series of electronic charges, comprising electrons, that sit on the gates of millions of transistors.

But this isn’t right either.

What happens when we copy the ‘software’ from one computer to another and run the software on the second computer? At first it seems we now have two copies of the software, but do we really have two copies of the software, or merely two physical representations of the same software?

After all, given that the copy of the ‘software’ exists inside the memory of the second computer (which could be in a different country) then it must be represented by a different set of electrons.

Because none of the electrons that represented the original version of the software form part of the copy (the original electrons are still on the original computer) we can be sure that whatever the software really is, it has nothing to do with electrons.

There’s more...

Even when the software is executed on the original computer the electrons that represent a given instruction in the computer’s memory are not the same ones that are used to represent the software when that instruction is executed in the microprocessor.

Again, we see that the software is different to how it is represented.

So what, exactly, is software?

The answer is that software does not exist – at all:

Software is simply a human intellectual construct, just like mathematics.

Software can only be represented in our physical reality, say by projecting something on a computer screen, printing characters on a paper page, storing a set of electronic charges in a memory chip. Or even as oranges arranged in a particular way in a park.

This rather abstract finding is important for the following reason:

If software is just a human intellectual construct then this means that machine intelligence must be a derivative of human intelligence.

Machine intelligence is where human intellect is repurposed in a way that allows it to exist within a computer, rather than a biological brain.

By necessity, machine intelligence must be very different to human intelligence:

• we are representing it using an imperfect intellectual construct (software)
• we are enhancing certain aspects (by taking advantage of the computer’s ability to process large volumes of data)
• we are ignoring certain aspects completely (like emotion etc.)
• we are then constraining it further by it by realising it in the form of multiple, wholly independent intelligence domains (different AI systems operate within their own closed domains)
• we are further restricting it by not providing any meaningful ability to interact with the surrounding environment

There is no way that machine intelligence is remotely like human intelligence in nature – even though it is derived from it, and – in a very restricted sense– can exhibit a degree of autonomous action or ‘free will’.

But, nevertheless, the nature of the intelligence we are creating in machines – while very different to ours – is absolutely as real as ours.

What is currently happening in the field of AI is that scientists and engineers are working with experts that have deep domain knowledge to aggregate, magnify and then commoditise their collective intelligence.

Because of the intellectual complexity of this task things are proceeding slowly on a domain-by-domain basis, but the end point is clear:

AI systems will eventually have been developed for all human intellectual domains.

We can easily imagine that these systems will be connected together into a network – rather like the internet – but where questions and answers will be relayed between the nodes, instead of data packets.

The resulting level of collective machine intelligence will be greater than the collective human intelligence that was required to build the network. This will be because the intellect of each ‘AI node’ will already be supra-human while network effects will mean that the collective intelligence of the whole network will be exponentially greater than the sum of the individual intelligences.

Even if some aspects of human intelligence – such as emotion, worry and irony, or even sentience as we understand it - remain absent from the ‘AI web’ then the resulting structure will still be extremely powerful, or ‘superintelligent’.

As soon as we realise that machine intelligence is simply a derivative, superior form of human intelligence then the inevitability of superintelligence is clear, even if the timescales and implications are not.

 

Ends