Page 35 - 10 Most Innovative AR VR Startups In 2019 up
P. 35
cucumbers with astounding We accept machines that act like if you have infinite disk space, the
accuracy. The problem is that their machines, but not the ones that process is expensive.
supervisors - the machine learning do the human stuff, like talking,
engineers or data scientists - don’t smiling, singing or painting. Of If you plan to use personal data,
know exactly how they do it. The course, this may change with time, you will probably face additional
problem is called a black box. as new generations grow up in a challenges. People around the
digital environment, where they world are more and more aware of
Artificial Intelligence supervisors interact with robots and algorithms. the importance of protecting their
understand the input (the data that privacy. They may be unwilling
the algorithm analyses) and the Talent deficit to share them with you or issue
output (the decision it makes). a formal complaint if when they
While the engineers are able to Although many people are realize you did it, even if you
understand how a single prediction attracted to the machine learning obtained all they gave you their
was made, it is very difficult to industry, there are still very few consent.
understand how the whole model specialists that can develop this
works. technology. A good data scientist Personal data and big data
who understands machine activities have also become more
Some AI researchers, agree with learning hardly ever has sufficient difficult, risky and costly with the
Google’s Ali Rahimi, who claims knowledge of software engineeri introduction of new regulations
that machine learning has recently ng. protecting personal data, such as
become a new form of “alchemy”, the famous European General Data
and the entire field has become Data is not free at all Protection Regulation.
a black box. It is a significant
obstacle in the development As I mentioned above, to train a The technology is very young
machine learning model, you need
of other AI applications like big sets of data. It may seem that Once again, from the outside, it
medicine, driverless cars, or looks like a fairytale. The biggest
automatic assessment of credit it’s not a problem anymore, since tech corporations are spending
rating. What if an algorithm’s everyone can afford to store and money on open source frameworks
diagnosis is wrong? How will process petabytes of information. for everyone. The Alphabet Inc.
While storage may be cheap, it
a car manufacturer explain the requires time to collect a sufficient (former Google) offers Tensor
behavior of the autopilot when Flow, while Microsoft cooperates
a fatal accident happens? How amount of data. Moreover, buying with Facebook developing Open
will a bank answer a customer’s ready sets of data is expensive. Neural Network Exchange
complaint? There are also problems of a (ONNX). These systems are
different nature. Preparing data for powered by data provided by
The black box is a challenge for
in-app recommendation services. algorithm training is a complicated business and individual users all
It turns out that web application process. Here’s an interesting post around the world.
users feel more comfortable when on how it is done. You need to You need time to achieve any
they know more or less how the know what problem you want your satisfying results and planning is
algorithm to solve, because you
automatic suggestions work. That will need to plan classification, difficult
is why many big data companies,
like Netflix, reveal some of their clustering, regression, and ranking Traditional enterprise software
trade secrets. ahead. You need to establish development is pretty
data collection mechanisms and straightforward. You have your
The research shows artificial consistent formatting. Then you business goals, functionalities,
intelligence usually causes fear and have to reduce data with attribute choose technology to build it, and
other negative emotions in people. sampling, record sampling, assume it will take some months to
People are afraid of an object or aggregating. You need to release a working version.
looking and behaving “almost decompose the data and rescale it.
like a human.” The phenomena is It is a complex task that requires In machine learning development
called “uncanny valley”. skilled engineers and time. So even has more layers.
December 2019 35
