The impact of Deep Learning on our lives

The last meeting of Science Café Cleveland had as its presenter Wyatt Newman, a professor at Case Western Reserve University, who gave a fascinating presentation on the state of Deep Learning, the term given by the Artificial Intelligence community to the next stage of AI development, where machines learn to identify things and make decisions for novel situations that they have not been previously programed to deal with. It is this feature, for example, that enables self-driving vehicles to identify the various things it encounters on roads and take appropriate actions.

He spoke about the dramatic advances that have been made based on our better understanding of how neural networks work in animal brains and the increasing power of processors to store huge databases and analyze that information and generalize from it. While we already know that machines are far better than humans at doing rote tasks, he said that there is no doubt that machines will be increasingly able to do cognitive tasks that we think are reserved for human intelligence, better than humans can. It is only a matter of time. One big advantage that machines have is that, unlike humans where each person has to learn anew what their ancestors knew, with machines that knowledge can be transferred intact to new machines with no loss.

There was of course the question of what would happen to all the people displaced by machines. He said that the new AI technology in undoubtedly going to cause great disruptions in society but the precise form it would take is less predictable. For example, with the advent of ATMs in banks, people feared the loss of bank jobs due to tellers being replaced. In reality, bank employment increased, it was just that new kinds of work emerged that were not as routine.

Another frequently quoted case is long haul truck drivers. He said that driving long distances on highways is the kind of low-hanging fruit that self-driving vehicles would first replace. He said that currently in the US, these drivers are limited to 10 hours driving per day. Even this is really too much because doing so day after day takes a real toll on drivers. It is too grueling for most drivers and most truck accidents are caused by either driver boredom or falling asleep or distractions. Automated trucks would be safer on the highways. But it would be advisable to have human drivers waiting at the highway exits when the trucks get close their destination, to take over and navigate busy city streets. It would like harbor pilots who are well aware of the local situation taking over from a ship’s captain when the ship enters port. So a new set-up would have to be created to take drivers to exits to meet trucks and to take them home from the exits once they release the trucks for the next trip on the highway. The advantage for truck drivers playing this new role would be that they would be driving for shorter periods and would now be always working close to home, like most other jobs, instead of the present situation where they are away for weeks at a time.

In addition, with self-driving trucks, they would not be idle for the 14 hours per day that the law requires their drivers to rest. Hence they could be on the road much more and so even with the same number of trucks, there would actually be more trucks arriving at and leaving exits and requiring human drivers to take over. So the overall impact on the number of truck driving jobs is not clear although the nature of the jobs will definitely change.

But what is clear is that these advanced robots are definitely in our future and going to have a major impact on our lives at work and at home and we have to brace ourselves to mitigate the adverse effects of the changes.


  1. anat says

    I wonder what those less routine jobs in banking are. When I go to the bank I always find a bored teller trying very hard to be helpful. And I recall reading somewhere that many bank employees are in need of some kind of economic assistance.

  2. Jenora Feuer says

    Speaking as someone whose father was a banker (admittedly in Canada, where some of the specifics are different), most of the less routine jobs already existed in one form or another, such as:
    -- Loan and mortgage handling
    -- Investment banking (IRA, RRSP, 401k, etc)
    -- Financial advice
    Basically, as ATMs spread out everywhere (and my understanding is that happened even faster in Canada, due to the more concentrated banking industry: by 1986 the Interac network was nationwide and you could take money out of any bank’s ATM, no matter which bank your account was with) the banks responded by slowing the hiring of new people, encouraging front-line people to ‘train up’ to higher level jobs, and allowing the level of employment to reduce naturally as people retired.

    And, of course, there’s still a need for front-line tellers for services that ATMs don’t provide, as well as customers who don’t like to change; just not as many tellers as before.

    Basically, anybody in the banking industry could see this was coming, and if you took the proper steps to mitigate what was happening, you didn’t have to fire anybody. That said, there’s no guarantee that everybody in the banking industry took the proper steps.

  3. jrkrideau says

    @ 2 Jenora Feuer
    When I hear that bank employment increased, I remember that so did the population.

    Of course, the banks seem to have a lot more marketing staff (as noted by CBC Marketplace) working in situations that make a 19th C Montreal sweatshop look good. This is good?

    Bring on the Guaranteed Annual Income!

  4. John Morales says

    Deep learning is just a technique. Matrices and gradient descents.

    Essentially, you could replace the term with “computer(s)” without losing meaning, since that’s how it’s implemented.

  5. Jenora Feuer says

    @3 jrkideau:
    I never said employment increased, I just said that they didn’t need to fire swaths of people because of the increasing automation.

    My father ended up taking early retirement nearly twenty years ago:he was only a few years from retirement anyway, he couldn’t be promoted anywhere but to head office but there’s no point in promoting someone to head office if they’re likely to retire before actually getting up to speed with the new position, and him being the manager of one of the busiest branches meant other local people couldn’t be promoted to the position he was in. So the bank offered him full pension if he retired early, and he took it.

    Never said marketing jobs being treated like sweatshops were good either. I’ve had friends call it pretty soul-destroying work.

    And, honestly, fully agreed on the Guaranteed Annual Income end of things. The idea that some people ‘deserve’ to be starving and poor needs to die.

  6. Canadian Steve says

    I think the guaranteed annual income, or some variation thereof is going to be a necessity because the increase in automation generally allowed for greater consumption in a limitless growth economy. More consumption of more different products allowed new industries to grow and hire displaced labour. But at some point this too reaches a limit, as each person can only consume so much and environmental limits are being reached as well. So I think the labour market is likely to get much worse in the next 20 or so years. Secondarily, once automation reaches sufficient levels it may be possible to produce enough goods and services for a non-poverty level of living without needing most people to work more than a few hours per week. (Imagine those truck pilot jobs were just 2 days with 5 off…) The big challenge becomes rethinking how we think about work and the economy generally, especially ownership of the robotics/neural networks that will do so much of this work.

  7. Canadian Steve says

    Should really reread for clarity before clicking post….

    I mean to say in that first part that I think the economic system we currently have will not be able to deal with displaced workers from this AI revolution as well as the economy redistributed labour. Even now we are seeing a massive increase in inequality based on the owners of the technology vs those who must work to survive. I think this will likely result in huge problems.

  8. KG says

    Deep learning is just a technique. Matrices and gradient descents. -- John Morales@4

    Here’s me thinking Geoff Hinton and his colleagues developed it over decades of intensive research, and all the time, John Morales could have told them how trivial it was!

Leave a Reply

Your email address will not be published. Required fields are marked *