TechSpot giveaway: Win a DJI Spark drone & a Star Wars BB-8 droid

Opinion: The business challenges of artificial intelligence

By Julio Franco · 5 replies
Oct 3, 2017
Post New Reply
  1. From a pure technology perspective, it’s hard to imagine something much hotter than artificial intelligence, or AI. Everywhere you look, companies of all types and sizes are talking about their AI initiatives and all the amazing things those projects will enable. Similarly, it’s nearly impossible to read recent stories in both the tech and general press that don’t mention AI.

    While I have no doubt that, technologically, AI is a fascinating new area of development that’s bound to drive some incredible new innovations, I am starting to have some doubts about the business opportunity for AI.

    In many ways, the business challenges for AI are similar to those that arguably still face “big data.” First, it’s hard to do, and the number of people with the skills to really create AI algorithms and other software is very limited, making the cost of creating any products and services with the right people very expensive.

    While I have no doubt that AI is a fascinating new area of development...I am starting to have some doubts about the business opportunity for AI.

    Second, the end results of the effort can be hard to quantify from a return-on-investment (ROI) perspective. In some cases, it’s easy to point to clear monetary savings or revenue increases from the results of either big data analysis or AI-driven outputs, but in a very large percentage of cases, it’s not. Sometimes it’s a process improvement that comes from the work—often a positive development, but not necessarily one that’s easy to associate with a dollar figure.

    The third and most fundamental challenge for many AI applications is with the inherent nature of what they’re designed to do. In essence, if an AI project is done properly, it should basically put itself out of a job.

    Let me explain. In many instances, AI is applied to a set of data in order to determine hidden patterns, more efficient ways of achieving/doing something, or just making a process easier or more natural. If the technology is applied in an intelligent way, then the results will be an improved process that’s cheaper, faster, easier or generally better than the manner with which it was done previously.

    That’s great, but often that discovery process only has to be done once. So, once you’ve figured out a better way to do something, the project is done. It’s essentially a one and done. Yes, there can often be iterative improvements made after the fact, but it’s often a case of diminishing returns. You might make 95% of the possible improvements thanks to that first round of AI-driven analysis, but then only make very minor improvements after that.

    From a business perspective, that’s clearly a challenge, because most tech-related business models are built around a continuous, ongoing stream of revenue and not just a one-time sale. You can certainly build successful businesses based on a single sale/project, but it’s definitely more challenging, especially because many of the efforts necessary to build a strong AI algorithm for a particular application are very unique to that project. As a result, it may be difficult to leverage that work across different projects/applications, which would be critical for building an ongoing, viable business.

    AI increasingly looks like an “ingredient” technology that could be challenging to monetize on its own.

    To be clear, there are certainly applications for which a constant flow of AI-driven analysis is essential—keeping an autonomous car driving for example—and those types of applications won’t necessarily face the business model challenges of other AI approaches. Even in these cases, however, there will likely be challenges in monetizing ongoing services, because the AI-driven features are going to be sold as part of a given device or service.

    Process-driven services are likely the best opportunity for AI from a business perspective, but it’s not exactly clear what those will be, how they will work, how they will be marketed, and if people will be able to understand the value in them.

    The bottom line is that, while there are clearly great opportunities to build some amazing AI technologies, the manner with which money can be made from those technologies isn’t quite as clear. Providing new levels of capabilities, improving how goods are designed, manufactured, sold and used, and generally increasing the overall “expertise” built into other products is clearly a noble goal, but AI increasingly looks like an “ingredient” technology that could be challenging to monetize on its own.

    Bob O’Donnell is the founder and chief analyst of TECHnalysis Research, LLC a technology consulting and market research firm. You can follow him on Twitter . This article was originally published on Tech.pinions.

    Permalink to story.

     
  2. jobeard

    jobeard TS Ambassador Posts: 10,709   +858

    Having worked with the Prolog language (symbolic processing AI language), there are some interesting facets to the coding and the solutions it (Prolog at least) provides:
    • logic can be coded to be deterministic(find A solution) or non-deterministic (find many solutions) and possibly all solutions
    • either way, solution(s) are not necessarily 'best fit', just logically correct (issue: criteria for best)
    • 'facts' are easily added to enhance or alter the derived solutions
    If you use GPS maps, you're already using AI. For fun, pick a destination you know well and map/get directions to it. Now while enroute, intentionally take a turn for an alternative route you like - - say back roads instead of hiways - - the mapping tool SHOULD recalculate a new route from wherever you are at the moment :grin:
     
  3. amstech

    amstech IT Overlord Posts: 1,836   +994

    No AI humanity has created thus far remotely resembles what I would call AI, atleast in my definition.
    And no, I don't think it needs to be near human brain scales of complexity to qualify as AI, but at this point the most impressive AI I've seen still makes a 5 year old look like Einstein.
    The ability to store information or speak does not make one intelligent or most importantly, sentient... capable of learning or making decisions. It will be, literally, centuries before we achieve what I would call true artificial intelligence.

    Start building the T1000's, just to be safe.
     
  4. jobeard

    jobeard TS Ambassador Posts: 10,709   +858

    @amstech sentient:

    adjective: sentient
    able to perceive or feel things​

    Abstract reasoning and creative activities are certainly outside the realm of AI.
     
  5. MilwaukeeMike

    MilwaukeeMike TS Evangelist Posts: 2,867   +1,203

    I think this is backward - and therefore not true. AI won't help us find improvements that people can't find, it'll do repeatable jobs that people can already do in a faster, cheaper way.

    Imagine if you worked at a record company and your job was to listen to new music all day and determine the artists that could one day be good. A computer could hear 100 songs at once, that wouldn't put itself out of a job, it would put the person out of a job.

    Robotics replaced humans to do simple physical tasks. AI will replace humans to do simple mental tasks.

    I'm not worried about this causing unemployment - it'll free up those people to work on other problems. And unless humanity runs out of problems to solve - we won't be out of jobs. And I don't mean problems like traffic and heathcare - I mean problems like 'it's too hard to buy a car online' or 'I can't find a hotel at the last minute in the city I just arrived in'
     
  6. OutlawCecil

    OutlawCecil TS Addict Posts: 201   +96

    As amstech basically was saying... It's easy to make something look intelligent. (which is as far as we've gotten) It's much more difficult to make something actually intelligent enough to learn for itself which is what everybody pictures REAL AI being.
     

Similar Topics

Add New Comment

You need to be a member to leave a comment. Join thousands of tech enthusiasts and participate.
TechSpot Account You may also...