Now and then, we hear of newly introduced AI tech with exciting features and updates, most of which sound almost too good to be true. Well, you know what they say about anything that sounds too good to be true – it probably is!
Rationally speaking, who does the actual checks and balances on these things? How can we rely on “statistics” and “research,” mostly drawn from very subjective data pools? Who is to validate some of these results from AI-producing brands?
There are many grey areas for expectation vs. reality in the AI world; what is promised is not always what materializes.
One could blame PR teams for the self-aggrandizing press releases for these new techs. Still, branding excellence or the advantage of selling vaguely described features and specifications to an averagely ignorant consumer population. And so, they cash out every time.
These AI brands cash out big time, riding on the wave of the trend of an exciting ‘all capable’ technology. And this is why you should be skeptical, especially where hyped-up AI is concerned.
The first red flag you should note is when advertisers and marketers start promoting new AI, is a shady demo, or as is the case with most products, the absence of one.
Realistically speaking, if a dude walked up to me, advertising a newly launched (or about to be launched) AI, and said something like, “This new technology is capable of reading 1000 words more than the average human per second.” I would be expecting a demonstration of some sort, at least to back up said capabilities. Talk is cheap, and anybody could come up with any ‘statistics,’ but can they walk the talk?
One could assume that before the product leaves the beta stage and proceeds into the production and distribution stage, a demo must have been presented to stakeholders and financiers who may have examined and approved its qualities based on visual representation. However, these stakeholders may not exactly be ideal experts who would tell one tech feature apart from another.
Even if there are experts present among the stakeholders, these experts do not even remotely represent the larger percentage of the target consumer population. This then implies that the average target consumer has to bank on the producers’ good integrity, take their word for it, and believe that what they say about their AI is true.
Another pointer to the fact that hyped-up AI is probably not all that it is said to be may be attributed to the general inability to fully infuse some of these features to other applications in a more generic manner. Going back to the Reading AI illustration, if the AI can ‘read’ but cannot interpret data as emotionally accurate as a human, doesn’t it just cancel out its reliability?
Can the AI fully comprehend all data precisely (or better than) the way humans would, and would it be able to read and comprehend innuendos, idiomatic expressions, and contextual expressions or jokes the way we would? I think not!
Who are these average humans?
Another reason you should be skeptical when it comes to new, hyped-up AI – as earlier hinted – is the blurry, unclear statistics that usually accompany the press releases which precede their eventual launches. When companies promise certain features and specifications to be expected from their AI, they put out information like “Statistics say that this tech can do xyz better than the average human”, the million-dollar question succeeding the statement should be “Well, who are these ‘average’ humans?”
I mean, most of these studies and research are not exactly carried out using the best of the pack. Suppose the population study is gotten from the typical unmotivated test team. In that case, the output will most likely fail to measure up to whatever prospects the AI offers. When the AI delivers as predicted, it gives the false impression that the tech company’s brand promotion team did well on their promise, yielding huge returns.
What the regular prospective client/customer fails to – or is yet to – comprehend is that the same results may not be as impressive if an actual unbiased study (or trial) was done with reliable human professionals. A human who is an expert at a particular task would most likely give the hyped-up AI a good run for their money, making said AI look mediocre.
You get to realize that in the real sense of things, a lot of these over-hyped AI are equal in capacity to the expert human’s production capacity. It then becomes obvious that the progress that some AI companies record on their product’s ability to perform specific tasks does not particularly translate to general progress.
To think of it, how do we gauge progress made in, say, for instance, chat-bots response features? Isn’t it by weighing the simultaneous progress made in other areas like the chatbot’s ability to accurately tell the user’s intended meaning is; especially in terms of figurative expressions?
If the AI fails to meet up to that expectation, then it goes without saying that it probably isn’t all that. Just exaggerated regular tech slightly more improved than its predecessors.
In the end, one can only boast about a products’ success or features if it meets up to and exceeds expectation. PR, marketing, and advertising teams will inevitably exaggerate their products’ competence. However, some panegyrics need to be questioned with rational contextual questions like ‘how resistant to failure is this AI?
And perchance the context in which it is used changes, how reliably effective does it remain? If it fails to adjust to required changes and maintain functionality. If it can only perform specific tasks and crumbles when its duties extend to more generic functions.
In truth, if an adequate, generally acceptable accompanying demo isn’t released to back up the released AI’s hype, then maybe you should scrutinize its credibility.