back to article Enterprises are rolling out more AI – to 'middling results'

Many organizations are struggling with artificial intelligence deployments despite believing that AI will be critical to business success over the next five years, according to a report by Deloitte. The 5th Edition of Deloitte's State of AI in the Enterprise report is based on a survey of 2,620 business leaders from …

  1. jake Silver badge

    Middling?

    From what I've seen it's more muddling. In several meanings of the term. Sometimes concurrently.

  2. Binraider Silver badge

    1) Collecting data is expensive.

    2) Classifying data, well, requires human eye to "train" data, usually in the form of an expert on that data set

    3) Errors have to be cleaned up to remove outliers in your dataset

    4) Weak, historical database designs, ancient paper records that won't go away, and/or spreadsheet reliance complicate the job of relating one piece of information to another. The role of an AI programme isn't to redesign your data hierarchy; and attempting to do so will "break" various other business functions without a very bottom up re-write of "everything".

    5) Employing 20 people to develop an expert system to "replace" the job of an actual expert on subject matter in question? Vastly more expensive.

    and last, but not least

    6) Having done all that farce with your AI box, what are you going to do with the output? (Relates to lack of experts).

    I speak from a position of "not for want of trying". I have ran several R&D programmes over the years to try and find something useful to do with AI in the absence of large teams of experts to draw on. The only ones that consistently come up with anything usable are in the image recognition space.

    1. Anonymous Coward
      Anonymous Coward

      Totally agree. I've worked in both public and private sectors where "AI" makes for exciting and promotion-inducing PowerPoint slides for executives, but the less sexy slides of the underlying data infrastructure to enable it are skipped over entirely, leaving data scientists with the 2 circles and then the owl meme when their group fails to deliver. The cultural shift to funding and creating machine-readable, re-useable data across organisations remains the greatest challenge of "industry 4.0", and in many cases would add value in itself just by enabling department heads to see it for the first time.

      1. Binraider Silver badge

        Many infrastructure organisations are reliant on paper drawings that pre-date WW2. 75+ years accumulated pencil markings. Going back and digitising all of that is to say the least, impractical. Adding instrumentation and condition monitoring to it is an equally ludicrous proposition to do in bulk.

        There are some tools out there that are incredibly potent; LIDAR being the most obvious one. But asking AI to work with the output is difficult. Is that really an obstruction you're seeing in your capture, or just a cow that's wandered somewhere interfering with your scan of the surface?

        Crossrail had a team of 90+ people working on it's BIM programme. That was a relatively small project in the context of just how much is installed in the UK. And despite the tools, Crossrail got into some very basic supply chain and project management / sequencing difficulties - more or less like every other major construction programme.

        Multiply that up by multiple asset owners that aren't obligated to pass information between each other. Some of them have active reasons not to reveal their dirty laundry too.

  3. b0llchit Silver badge
    Devil

    Recursive AI

    ...that many organizations are struggling with "middling results"...

    They should put the success/fail modalities in an AI and predict whether they will succeed or fail. Then they have a successful AI to predict their downfall?

    1. Strahd Ivarius Silver badge
      Devil

      Re: Recursive AI

      The first rule being: if you use Excel, you fail

      1. ArrZarr Silver badge
        Windows

        Re: Recursive AI

        Excel is a wonderful tool for datasets up to about 200k rows long (as long as the processing is light)

        That being said, you need to know what you're doing to get anything useful and reliable out of it.

        We use it for dashboarding extensively and the flexibility of being able to define the exact formula for a cell makes it vastly superior to DataStudio or Tableau for data sets in our ranges.

        Google sheets realistically caps out at 100k rows with very light processing or 30k rows with heavy processing.

        1. Binraider Silver badge

          Re: Recursive AI

          Correction : It's a useful tool for analysing datasets as you describe.

          The inevitable mistake is that the spreadsheet becomes the master, and you lose all referential integrity between systems and processes at that point.

          I have seen this in every single industry I've ever worked in.

  4. Tomato42
    Trollface

    What? Putting square pegs into round holes doesn't bring business value?! Inconceivable!

    1. b0llchit Silver badge
      Go

      Sure you can! Just apply enough pressure.

      1. vtcodger Silver badge

        "Just apply enough pressure."

        If you can't force it, you need a bigger hammer?

        1. b0llchit Silver badge
          Joke

          You can use a bigger hammer, yes.

          However, it turns out that the casual user has all the tools required. A casual user is able to fit any connector to any socket. These casual users have perfected the many-to-many connector system. The technology doesn't work, but the connectors fit.

          So, please go ahead and use a casual user.

          1. Anonymous Coward
            Anonymous Coward

            If that fails, try the nearest two year old. They have a knack for inserting the most "interesting" objects into electronics jacks and wall sockets...

            My personal favorite was a 3 year old who inserted a peanut butter and jam sandwich into the VCR and pressed "play" to watch her cartoons. *LOL*

  5. Tron

    In other news, the Emperor has some new clothes.

    AI is a con. A sticker attached to anything.

    The real thing is inherently unreliable. It learns by making mistakes and works on historical/legacy/out-of-date data. You wouldn't want to replace competent staff with it.

    AI is proof that most corporate users have no tech I.

  6. xyz Silver badge

    If one more person mentions AI...

    thinking I'll buy into whatever crap they're selling I'll slap them. At least they've stopped trying to blockchain everything which was the last marketing fad that went phut.

    Also "artificial intelligence" is a term that I find racist and smacks of slavery. Ask anyone in 400 years time.

  7. Anonymous Coward
    Anonymous Coward

    AI. Always though of it as ass insurance.

  8. Big_Boomer Silver badge

    Awfully Inadequate

    That's my take on AI. Not one of these AI systems are artificially intelligent. At best they are neural networks that have learned to take a trained/educated guess at something and have a high-ish chance of being correct. However, they are often incorrect and therefore utterly useless if the data is to have any value at all. I've seen a lot of AI hype over the last few years and I still maintain that at best what they are selling are Expert Systems and that there is ZERO Intelligence in any of their systems. It reminds me of all those silly psychometric tests that HR types thought were the dogs-danglies until someone actually researched their effect and it turned out that they were utterly useless and it would have been just as effective to just bin every other CV. These Expert Systems do seem to be pretty good at pattern recognition and can process data WAY faster than humans, but the positives still need that trained human eye to confirm that they are correct and THAT is where the Intelligence comes in. One day we may well have Artificial Intelligence but not for some time yet.

  9. Eclectic Man Silver badge

    I want

    An AI system that detects and reports biasses in the training data. For example, in a legal setting an AI would detect whether some factors (wealth, family connections, race, sex, sexuality, age, etc.) were correlated with outcomes when irrelevant to the actual offence. But I suspect that might be bit 'unpopular' with the 'establishment'.

    I suspect that one of the problems with AI systems is that they are built by people who are enthusiastic about AI, without cynical old gits like me having any input to the design, purposes, 'training data sets' or anything else. Rather like the international data processing network procurement manager who phones up the the security consultant one day before 'go live' to get their formal approval.

    I do take one issue with the article. If you don't have to re-design your company working to take advantage of (an effective) AI, then your AI really isn't that revolutionary, is it?

    1. Anonymous Coward
      Anonymous Coward

      Re: I want

      I'll raise your "bias detection" system a "biased bias detection system"...

      1. Eclectic Man Silver badge

        Re: I want

        Upvote for your 'biased bias detection system', but bemused by the downvote (probably one of my stalkers).

  10. Anonymous Coward
    Coat

    Reuse

    Deloitte could save themselves a lot of time and effort by releasing the same report but replacing "AI" with "blockchain" and then with "quantum".

    1. Anonymous Coward
      Anonymous Coward

      Re: Reuse

      No fair! You have to pay for a subscription to get those reports... :P

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like