Speech impediments and predicting the unpredictable

My favourite Economist section, The Technology Quarterly recently came out with a focus on voice recognition.

The key reason for the focus is that Microsoft Switchboard reached a 5.9% error rate – which is the same error rate as a human transcriber and considerably better than the average offshore call centre operator I’ll bet…

Speech Impediments

There are investment outcomes from this over the next few years I’m sure, mostly in continued cost savings for companies, lost jobs in call centres and more wage pressure at the low end of the income scale. But these trends have been in place already for a number of years.

What’s of more interest to me is the exponential improvement in accuracy over the last five years.

One of my key “investment pivots” is the take up of driverless cars – knowing whether they will be here in five years or fifty years is incredibly important for a range of stocks, and so I’m looking for any clue I can find in this area.

In my mind, the problems that cars have to solve to become driverless are very similar to the problems that speech recognition has to solve: here is a pattern that could be a range of things, work out what the pattern really is.

I’ve had a decent play with machine learning from the investment side, and the problems that you practice on for machine learning are frequently processing images or sound – once you have converted images or sound into data, the models you use are often identical.

So,  the lesson I’m taking is that machine learning has taken a mini-leap forward.

Battery costs also took a mini-leap forward last year:


Combined, these mean that regardless of whether you are in the “driverless cars are 50 years away” camp or the “driverless cars are just around the corner” camp, the time frame looks to be sooner than you would have thought a year ago.





  1. Damien. How can there be “investment pivots” around a trend that is systematically replacing human labour with machine labour? As more and more people become an unnecessary component in the machinations of our society, the notion of value and “investment” becomes decoupled from reality. A jobless person is a passive appendage to the machine society. The self replicating robotic manufactories can make anything we want, ultimately at will. Are we hoping that the MegaCities will hold together against all odds and we can rely on a 19th century simulacrum of “Father Knows Best” and “Leave it to Beaver” work-a-day lives for the masses? Fake jobs, fake work ethics, fake economy so that the Bankers and Finance sector can trade money between themselves like a souped up 21st century online game of Monopoly?

    I would really like someone to explain where the faith in our financial system comes from. How can economists talk about decommissioning humans and at the same time believe that there will be “investment” and “growth”? Worse still is the obvious fact that in most cases of “investment” we are actually talking about fraud, inefficiency and corruption. “Investment” has actually become a synonym for bad behaviour of the elites. “Investment” now usually means destroying someone else’s future for a bit more snout in the trough

    Banks – corrupt, parasitic and unproductive.
    Media – ditto
    Housing Industry – ditto
    Legal system – ditto
    Medical drugs – ditto
    NBN – ditto
    CSIRO – actually useful, but a bad “investment” so they closed it down

    All of this financial bad behaviour is served up with a heaped helping of pious puffery from the economics professions. Short Bus 19th century mathematics groping for stocks and flows and equilibriums straight from the theatre of the absurd.

    So, tell us how it works. Convince us that there is some substance to “investing” in the 21st century, and it isn’t just another grubby grab for a few dollars courtesy of the boot on the other man’s neck.

    • It is obvious that robotics (be it through robot cats or robot factories or robot anything else) has the potential to improve material living standard (by making things, better, faster, cheaper, etc) than a human can. More services/goods for less . clear win.

      There question is who gets to capture the benefit from this increase in wealth. … if it’s one grubby plutocratic robot factory owner – that’s one outcome. If it’s the society of small people (who get to consume cheaper – and eventually free – food, medicine, transport, etc) – that’s another.

      There is nothing in robotics per se that points to one outcome or another. It is property laws (including intellectual property laws) that will decide this – and there is the investment opportunity… being able to capture some of the benefit by owning the right things. Trying to be the owner of the robot rather than the guy whose job the robot just took.

      • I want to object here to you insinuating that a robot factory owner would be grubby and plutocratic. I know it’s not your point but the association of manufacturing with “grubby” is genuinely a problem.

        Class revolution against factory owners is historically a bad thing so let’s not encourage that, especially when Australia needs more of them. A lot more.

        A factory with industrial robots is a huge investment on a very long time frame by people who intend to manage it properly, invest in worker training etc. This is not like property or share investing. Look at how hard it is to do manufacturing well. The sunk capital is enough to turn away most. Just imagine asking a bank for a million dollar equipment loan instead of a property loan. “Sir, one day I might make a profit and you won’t have to fire-sale all this equipment at scrap-metal prices”

        This kind of person is less likely to be grubby and plutocratic IMO.

      • Ranald, I don’t mean the factory owner. I mean someone (Company X? Monsanto? Krupp? Microsoft?) manages to come to own and control the key technology. And proceeds to extract every last penny from it, licensing it out just cheaply enough to make the majority of humans unemployed and impoverished, but expensively enough to make the other half still have to work 50 hours a week for their daily bread.

        Think of something like the pricks at Turing Pharmaceuticals…. https://www.google.com.au/amp/www.forbes.com/sites/luketimmerman/2015/09/23/a-timeline-of-the-turing-pharma-controversy/

      • If the masses do capture a large portion of the benefit – by goods and services becoming cheaper – this is certainly a good thing. But, there seems to be a serious problem in terms of government planning. Here I am mainly talking about countries like U.S., France, Japan, etc. (but Australia may also be affected). Many governments, and State pension funds, have been planning on the assumption that when the economy expands, it automatically leads to a corresponding increase in tax revenue and investment returns. They have promised ridiculously generous pensions and other entitlements into the future on that basis. But, it may happen that the economy does indeed expand (in the sense of us all getting access to more goods) but that very few jobs, taxes, share dividends are generated in the process. Governments need to plan for this radical shift in the shape of the economy, instead of trying to sail along with business-as-usual.

      • Correct. Robotics are a tool, and the outcome of advances in robotics will mainly be determined by the legislative changes that will accompany it.

        In terms of the original ‘investment’ question.. Investment in robotics and AI applications are far better than the majority alternative – investing in speculative bubbles. At least they are achieving growth by increased productivity, rather than greater population.

      • Jason (JC)

        I may have misread your post but I’m pretty sure just about every major technological advance has been associated with extreme speculative bubbles. Canals and railroads certainly were, as was IT more recently, and there were hundreds of IC car manufacturers in the first part of the 20th century but over time all but a handful ceased to exist.

        If Damian Klasson and MB are attempting to use the technology advance associated with driverless cars as an “investment pivot” then I guess they will have to factor in the need to navigate through lots of volatility.

    • It is obvious that robotics (be it through robot cats or robot factories or robot anything else) has the potential to improve material living standard (by making things, better, faster, cheaper, etc) than a human can. More services/goods for less . clear win.

      There question is who gets to capture the benefit from this increase in wealth. … if it’s one grubby plutocratic robot factory owner – that’s one outcome. If it’s the society of small people (who get to consume cheaper – and eventually free – food, medicine, transport, etc) – that’s another.

      There is nothing in robotics per se that points to one outcome or another. It is property laws (including intellectual property laws) that will decide this – and there is the investment opportunity… being able to capture some of the benefit by owning the right things.

      • +1 In the days of human/animal power 99.9% of the population lived in utter poverty. What we do with our tools depends on us or what we allow society to do with them.

      • A universal minimum income for All is the answer to stopping capitalist control of our lives, Peachy.

        This is a simple and profoundly useful economic tool to move towards equality that has been undermined by capitalists each time it has been tried in various countries.

      • Indeed. Recent decades suggest most of the productivity(ie: efficiency) gains will simply be pocketed by the rich. Some serious steps will be needed to stymie taxation avoidance by MNCs and fund a social wage.

    • I think humans have a lot to offer each other. If robots take away manufacturing jobs, then we’ll probably create more services jobs, and in the creative arts. Would you really want to see a play performed by robot actors?

    • This concern is a rounding error. The real issue is when do we get the first robot armies and “police”? The rebirth of tyranny in what is now the developed world.

      • There will be robot police as soon as a robot police officer can’t be fooled by a low IQ inveterate ice user petty crim relying on a few tips from his mates.
        At least a century away, I’d reckon. Robots struggle with humans acting rationally and sensibly who are essentially on the robot’s side – how are they going to cope with humans on everything but roller skates or with mental issues spouting total nonsense at them and/or deliberately trying to confuse them?

        The ‘Jess from JetStar’ scenario is an example of why.

      • The essence of riot policing is dealing with many human adversaries operating in a rules-free environment, often irrationally. This is still very difficult for AI if you’re after anything more subtle than just spraying the crowd with rubber bullets. If you’re happy with spraying the crowd, then, yeah, it’s probably technologically possible any time someone wants it enough.

      • That’s my point. Riot police and soldiers disobeying orders is one of the most important safety mechanisms against tyranny. Once you have robots in that role, you’re done.

      • I guess I’m still unconvinced that there is technology capable of controlling a crowd of unpredictable humans at all (far worse than herding cats!), let alone doing it without sparking additional protests against government brutality.

    • Damien. Welcome to Macrobusiness commentariat. Most of which despises business. Abhors profit not redistributed via Government. Urges higher taxation on those earning about average incomes and higher taxation on corporates of every size. Views those that work in finance, property, mining, etc as corrupt greedy bastards. Dystopian revellers near all.

      Fortunately for you and the embryonic fund, secretly they do want to make big moolah, drive flash cars and have a Harbour view.

      Good luck 😉

      • “Views those that work in finance, property, mining, etc as corrupt greedy bastards.”

        That view is the natural conclusion from reading years of posts from the owners of this blog! It is also a view that is largely self inflicted by the FIRE sector. Is the MB commentariate delusional about the nobbling of the FIRB? The excesses of the Banks? Etc. etc. etc.

        “Fortunately for you and the embryonic fund, secretly they do want to make big moolah, drive flash cars and have a Harbour view.”

        That is a very poor comment. Most cheaters and petty crooks are weak minded. What they do is seek to rationalise their own bad behaviour by claiming that everyone else is also immoral or slightly corrupt as well. From there they can avoid any effects of conscience based on the fact that everyone else is on the same level. The next step is to claim superiority based on the fact that they are simply being pragmatic and taking advantage of what everyone else would like to do, if they weren’t so stupid.

        There is nothing new in the foul rag and bone shop of the human soul. Having the morals of a rat is a very old story, and it starts with weak excuses. Do you really think that the people on this blog are just greedy little banker wannabes too timid to go out and grab some dirty money? If you do, then that probably says a lot about you.

      • Scratch below the surface and yes, most want to make a buck, live in a nice home and drive a cool car. They pretend they don’t. But they do. They’re only human.

        Eyes wide open.

      • The “MB commentariat” wants to raise taxes on people earning under a hundred grand ? Can’t say I recall seeing that.

      • I didnt get the memo that I was a “corrupt greedy bastard” having worked in finance since 2007 or so…

        Must have been when i was down at the Maserati dealership…..

        Good use of reductio ad absurdum Mike – you get 10 bonus anonymous internet points for that one.

      • Chris, that wasn’t directed at you.

        Damien arrives here as an new entity offering investment opportunity (as you do) along with a somewhat tentative initiation into the blogosphere. He is immediately met by a barrage of anti business anti everything comment along the lines I parody above.

        My point to Damien is ‘do not despair’, there are many at Macrobusiness prepared to ditch the tin foil hat and don the trilby so to speak. Happy to get get advice, make sound investments etc. Not everyone is anti business (but the loud voices here often are).

        I made good money during the mining boom, frankly, a lot of money. I admire the risk, the execution and the reward of a good deal.

        Pretty sure there are many that share my love of opportunity.

  2. Have you measured frustration of end users dealing with these ‘bots’, when trying to actually get an answer on anything?

    Also “5.9% error rate” hides all manner of sins. There is no measure of how big the error is. A human might get the odd word wrong or mis-spelled but be still in context, whereas the computer will be off in a different universe. Both are still recorded as an ‘error’.


  3. Just a small nit pick… ‘Voice recognition’ generally means identification of the speaker based on the unique characteristics of their voice (e.g. the ATO call centres now have this as an option for ‘proof-of-identity’… which is why I’m never calling them again). ‘Speech recognition’ is what you mean here: transcribing speech to text automatically.

    See paragraph 4 here: https://en.wikipedia.org/wiki/Speech_recognition

  4. Jetstar has a a charming assistant, Jess, who provides all types of links and generalized useless information. The way to make it work is to express dissatisfaction in foul language, then Jess is suspended, and a real human intervenes, using specific customer data. Those virtual assistants don’t like bad language.

  5. I’ve worked in software development since the early nineties and think we’re at a turning point with deep learning. A few think to look at include nvidias self driving car demo from CES


    Also the fact that Tesla is installing the hardware for full autonomous driving in all cars now. I think Tesla has forced all other major auto companies to invest heavily in this area. Tesla model 3 will start pumping out near the end of this year which will really start changing.

    But the real change will be when we start taking that new hardware and deep learning software and apply it to everything else. Any task that can be learned can be replaced.

    Last year I predicted that by 2025 that electric cars would outsell ICE. I don’t think that will be on current volume. More like crashing ICE sales and rising EV autonomous vehicle sales. By then I’m guessing I’d have sold both my cars and using an Uber like service.

    As for battery tech. Good enough now but not cheap enough. Probably only a matter of time before incremental cost improvements make it viable to go mostly off grid.

    • Self driving vehicles are a tougher nut to crack than most people think. Although I agree that when this happens a lot of other activities will follow, it remains to be seen when this will happen.

      • The Traveling Wilbur

        Disagree. I reckon we’ll see autonomous vehicles crack a lot of nuts. Soon. And frequently.

      • Can’t wait for the first cracked coconut lawsuit. Seriously though, the problem has always been the same. Technologists underestimating the difficulty of the “easy” stuff (bipedal locomotion, artificial intelligence, space travel) . As others have said, driving a vehicle in difficult conditions or around pedestrians and predicting their behaviour (as opposed to reacting to it) is not that easy, as anybody in the carbon world can attest. A fly is better at flying than a military drone, with a brain the size of a sugar crystal. Food for thought.

      • I’ve seen some of NVIDIA’s videos of self-driving cars. They look great, and apparently work in a variety of conditions, but I’ll be happier when I see self-driving car videos that aren’t part of someone’s advertising campaign, edited to make it look as impressive as possible.

    • Any task that can be learned can be replaced.

      Although I’ve only reached the foothills, and so far only have an interest in autoencoders, from what I’ve seen of others using deep learning, and learned in my attempts to apply it so far, I’d suggest that the use cases are signficantly more narrow than what you suggest. In particular, we are at least several major breakthroughs of approximately deep learning calibre ( a once in a a generation breakthrough) in their own right (and a suitable data set is yet to be developed) from the notion of an intelligent system ‘learning’ a human job in the same way that deep learning systems currently ‘learn’ to identify faces or common objects in images.

  6. Call centers won’t even be necessary if the companies allow you to do things online. They don’t. I imagibe a future where you ring up about a billing mistake, and the AI tells does not give you the option to dispute the fee. It will definitely ‘save money’ on call centers, although angry customers may burn the company down.

  7. The one thing I wonder about with driverless cars is that they will need to be impervious to hacking, or else they will become WMDs. But is there anything impervious to hacking?

  8. The people need to rise up sooner rather than later and kill the rich robot owners and take those robots for themselves! The longer it takes for this to happen the worse things will become!!

    • The Traveling Wilbur

      Uh… Reusa, hate to have to be the one to break this to you, but you want robots. They’ll give you more time to spend picking out new IPs and lower the maintenance costs on your existing ones.

      PS did you see the tragedy with the Health Minister? Being forced to apologise for buying an IP. And worse yet, pay back expenses. And a penalty?!! Criminal. Tony wouldn’t have dreamt of such things.

      • Of course she will be penalised: for getting caught, and drawing attention to something they’re likely all doing.

  9. Aside from Hansard reporters, court reporters and similar, what occupations would see a significant productivity gain from moving from untrained typist transcription speed of around 20 wpm to speech speed of around 100 wpm? For example, it has been said that the average developer produces on average around 10 lines of finished code per day (per the Mythical Man Month- YMMV, significantly) – even a terrible typist could knock that over in minutes.

    Now that there is a computer on every worker’s desk there aren’t that many largely typing jobs to eliminate.

    • Thats right. I lay the blame squarely on Microsoft for inventing the Ctrl+C, Ctrl+V…. our esteemed journalist at Fairfaux, NewsCorpse have already been very innovative and ditched typing altogether…. only Ctrl+C, Ctrl+V from now on.

  10. why so worried about automation? It will/has/is ultimately FAIL. Why? Human sabotage. E.g. whats wrong with global finance atm? robots that are running it, are sabotaged.

  11. Damien,
    I have worked in/on some of the state-of-the-art techniques and applications that you are seeing in today’s world (people/object recognition, scene understanding, etc). The advent of GPUs (cue NVIDIA) and deep learning has certainly helped push the technology to an extent that what was considered “plausible in 2050” few years ago, is now considered “possible in 2025”. Drones that are tracking “targets” from 30,000 ft are already in working prototype mode (i worked on them). Processing satellite imagery was a big ask few years ago, but not any more. Digital Globe Inc have gone to extreme lengths to get ITAR permission to enable them to sell these imagery to practically anyone. I know couple of companies that are “feeding economic intelligence” , extracted by processing satellite imagery daily, to Two Sigma and the likes. Autonomous vehicles is still a touchy issue in the consumer-vehicle space due to lots of concerns about “rogue behaviour” but not in military-space. But what the consumer-autonomous vehicle R&D will bring forward a lot of a autonomous capability in the other domains/space( remote surgery, remote servicing of machines, etc). My two cents. 🙂

    • Right, well, that rises some interesting questions, aside to the eternal: “what is our purpose after all?”

      Then, what do you trade when everything you have is valueless? Will life still have value? What is life – at that point – anyway?

      If we haven’t figured out our own purpose, then how do we expect that autonomous artificial intelligence released into the wild will spare its once-upon-a-time masters?

      Somehow I think that “robot with a hard-on” will be the least of our problems…

      Oh, and here’s your Futurama-based question of the day: vs lbh pna qngr ebobgf, jul vf mbbcuvyvn fgvyy gnobb?

      (Anyway… off to shoveling more road-base… robots doing my job, my foot!!)

      • The primary reason for living things killing other living things is: sustenance, territory/threat.

        Robots don’t need to eat us, in fact, we’re probably corrosive to them.
        We are not much threat to robots. By the time we feel threatened by them, we’ll be dependants, not masters. (How many people still remember anyone else’s phone number? I havent memorised a number in years)

        We won’t share the same territory as robots. As evolved creatures we are heavily dependent on Earth-like living conditions. Water, oxygen etc. We will in fact start dying without an appropriate gravity well. Robots on the other hand will not find space to be all that hostile. For their mega processors they wont want to be anywhere near a heat source. For their precious metals they’ll want to be near metallic asteroids where the metal is near/on the surface (whereas a big planet like earth the heavy stuff is in the center, with just a smattering on the crust).

        So rejoice humanoid! Your future robot masters will consider you to be a slower, dumber, toxic bit of wildlife that they needn’t show more than indifference to!

  12. I am interested in how automated cars will interact in the pedestrian space. If pedestrians know the car is automated, then it seems to me to be a reflexive process in that people will behave differently around them.

    Presently part of what stops pedestrians just stepping out is the risk that the driver won’t see you and run you over. However with an automated vehicle this shouldn’t happen, ergo pedestrians are more likely to just step out with the car programmed to yield. How do you get around this and stop pedestrians from simply taking over the roads?
    Well perhaps you program the car to not yield, but then haven’t you legally programmed something licensed to kill/injure?
    Then maybe you take it to the next level and have a car with sensors that can distinguish between pedestrians accidentally straying into the path of an automated car versus delierately doing so, and give them a license to injure the later and not the former. But that has to be a more complicated problem to solve.

    Currently the solution is to keep the human in the front seat, and if the computer can’t decide, it just shuts off and palms off the decision to the human who has to act as a blame/liability ‘acceptor’. Which is not automation at all to having something shit itself as the answer to a problem.
    The recent headless Uber outings in SFO were hillarious with cars running stop lights etc. Of course Uber claims it was the driver at fault, but they would say that, the other option to say their technology is deficient which goes against their vested interest in the matter.

    I guess the short term reality will be freeways/motorways with automated driving lanes, and then eventually the technology reaches suburban streets etc.
    But if your child was killed on a suburban street by a headless Uber, no appeals to ‘how safe automated cars are’ would appease the need for legal action. The question is who is going to cover that risk of liability financially?

    Surely not a patsy they decide to put in the front seat on minimum wage?

Leave a reply

You must be logged in to post a comment. Log in now