• Post Reply Bookmark Topic Watch Topic
  • New Topic
programming forums Java Mobile Certification Databases Caching Books Engineering Micro Controllers OS Languages Paradigms IDEs Build Tools Frameworks Application Servers Open Source This Site Careers Other all forums
this forum made possible by our volunteer staff, including ...
Marshals:
  • Campbell Ritchie
  • Paul Clapham
  • Jeanne Boyarsky
  • Junilu Lacar
  • Henry Wong
Sheriffs:
  • Ron McLeod
  • Devaka Cooray
  • Tim Cooke
Saloon Keepers:
  • Tim Moores
  • Stephan van Hulst
  • Frits Walraven
  • Tim Holloway
  • Carey Brown
Bartenders:
  • Piet Souris
  • salvin francis
  • fred rosenberger

Succeeding With AI

 
Ranch Hand
Posts: 460
2
  • Likes 1
  • Mark post as helpful
  • send pies
  • Quote
  • Report post to moderator
Hi,

I like the way you laid out the book . Something tells me you've seen some projects that should never have had AI applied to them.
What has been the most egregious use of AI that you've seen ?

thanks,
Paul
 
Author
Posts: 7
6
  • Mark post as helpful
  • send pies
  • Quote
  • Report post to moderator
It depends on how we define the worst use of AI:
  • Something that causes significant loss of money to a company?
  • Algorithm looking for the nail.
  • Copying use cases.
  • Highly public failure?
  • AI shouldn't have been used at all.
  • AI works, but you wish it didn’t.


  • To protect the guilty, I would have to stay general when we are talking about the actions of a single company I directly worked with. In other categories, I would use some public examples of failures (I am glad to report that I was not directly involved with those failures).

  • Loss of money - seen a lot of those - I even developed a "template of how to fail." Step one is to build much infrastructure without a clear idea of what analytics to do with it, sometimes masked by simple POC that is never monetized itself. After that investment is complete, the second step is to start a single analytical project, preferably as a copy of something that you have seen other people do. Step 3 - realize that you don't know how to apply those initial analytics in your business. What is especially sad is when this happens in the small company, or already financially struggling company.
  • Algorithm looking for the nail - I have seen cases in which technology is really cool, and developers of the cool POCs got a lot of experience with the technology. The problem is sometimes that there was no ability for a business to collect data that algorithm would need; in other cases, it was simply poor business cases with little interest in the marketplace. All the while, chances are that if you review the business side carefully, you will find a situation in which not much more then a simple statistical test on available data may give you an answer you need.
  • Copying use cases - not to pick up on sentiment analysis, but not every business benefits from the sentiment analysis - especially if your user and your buyer are different people. I give an example in my book in the form of "if you are selling traffic signs to the local government, public sentiment about traffic signs in the town is not likely to be an actionable decision for your enterprise."  
  • AI shouldn't have been used at all - according to MMC Venture's "State of AI 2019: Divergence," 1 in 12 European startups puts AI in the center of their value proposition. However, according to the same report, in only 60% of the cases, they were able to find that AI was material to the company's value proposition. I don't believe that this problem is present only in Europe, and that also brings a problem of overclaiming the use of AI. Strictly speaking, that also means that in some of the examples I give, you can say, "but that system didn't really use AI."
  • Highly public failure - there were quite a few of those. An AI bot got much attention on Twitter few years back for very wrong reasons (the name of the bot was Tay).
  • AI doing wrong things - recommendation engines that continue advertising products after you buy them are wasting user time and advertisers' money. Biases in the algorithms are a big problem when they are making impactful decisions, such as determine job prospects (or even worse sentencing guidelines). There were reports of the government system used to detect fraud with no sufficient checks done before initiating formal fraud accusation - and then subsequent reviews showed very high false-positive rates.
  • AI works, but it would be better if it didn't - AI may be able to diagnose some contagious illness but is of little practical use if it requires funneling people in the diagnostic apparatus that is scarce (compared to the prevalence of the illness). Even worse when it increases chance of illness propagating due to the diagnostic tests. Let’s not even mention all the possible misuses of the AI in the hands of the totalitarian governments.

  •  
    Veljko Krunic
    Author
    Posts: 7
    6
    • Mark post as helpful
    • send pies
    • Quote
    • Report post to moderator
    Also, thank you - I am glad that you liked the book.
     
    Marshal
    Posts: 69493
    277
    • Mark post as helpful
    • send pies
    • Quote
    • Report post to moderator
    What makes you think it is only totalitarian governments who misuse AI ?
     
    paul nisset
    Ranch Hand
    Posts: 460
    2
    • Mark post as helpful
    • send pies
    • Quote
    • Report post to moderator
    Thanks Veljko . The common theme I picked up on from your response is the importance of having a clearly defined use case and the problem you are trying to solve. I've found it is a recurring theme in IT in general not just AI.
     
    Oh. Hi guys! Look at this tiny ad:
    Devious Experiments for a Truly Passive Greenhouse!
    https://www.kickstarter.com/projects/paulwheaton/greenhouse-1
      Bookmark Topic Watch Topic
    • New Topic