salespeople Skynet level, reviews of the level of God / Habrahabr

September 12, 2017
While we look, Apple and Google are fighting for the creation of AI, and Mask and Hawking are afraid, somewhere near the artificial intelligence is literally entering our lives from the side, where we did not expect. As I, reading the well-known in relatively narrow circles website about venture business and other start-ups, did not expect to stumble upon a very quite advertising announcement of the hackatoon of the M-video trading network on the very not-so-anticipated from the store selling irons and plasmas: on the development of chat bots for retail and the use of artificial intelligence in retail . Being, to some extent, an enthusiast of the topic of AI, I was already namylilsya read what application he was going to find in retail, but even on the site of the Hakaton could not find any detailed description of the idea. It is only reported that ten teams will compete for prizes, the main one of which, I understand, is not even 150,000₽ as such, but getting from M-Video a contract for finalizing their idea before commercialization.


Hakaton's theme is formulated rather blurry: "Aggregation of reviews on the site" and "Assistant to the seller for the selection of goods in the store" – I did not find any details in the original article or on the website of the hackaton itself. From this it became only more interesting. What will the AI-assistant of the seller do to find the models he needs in the database, which often takes so much time in the store when you are interested in a particular razor model? Why, for example, does AI need to aggregate reviews on the site, when they are already and so rigidly cataloged? Only by digging a little Google on relevant requests, I found enough food for thought, which allowed my fantasies to disperse already in earnest.

It's not an attempt to guess what actually and how will be developed on the hackatoon "M-Video" – I'm not aware of their plans, there is no information on the hackaton website. This is a free essay on motives. What can be done, armed with AI to solve the tasks.

Aggregation of reviews on the site

On the topic of "reviews" and "AI" on the Internet are already gossiping some materials, on the grounds of reading which, I had two possible ideas:

Analysis of reviews with a division into positive and negative in automatic mode

Which store is not interested in seeing that under the product line there are always only positive reviews? Manual moderation of reviews is too expensive in man-hours and man-money, and any mechanisms of user self-control are not just in the sense of protecting against cheating an honest, but unpleasant response from dissatisfied consumers. And then the online store comes to artificial intelligence, which, analyzing the accumulated feedback base, will learn to separate positive responses from negative ones, no matter how Aesopian language the latter were written. By the way, a useful development would be: imagine Habr, who no longer suffers with experiments on fine tuning of self-regulation mechanisms, but simply cut the AI, which knows exactly when the post corresponds to the subject matter of the resource, even if it is in a corporate blog, and when it is an advertising bullshit or just chewing the cumin for the sake of easy pluses. Which will no longer be, when the AI ​​takes over. He simply decides what to show us in accordance with our same requests for what Habr is a cake. And perhaps one of the teams on the Hakaton will start writing something that, when adjusted for another task, will become a karma killer.

Aggregation of reviews on external sources

Even more rich topic, if you think that reviews on products are left not only on the sites of different online stores, but also in aggregators like "Yandex-Market", and in blogs, and in general, anywhere else. And if you recall Google's experiments with a neural network in an interpreter, there are generally some unprecedented prospects: once having covered the entire Internet, AI could collect all relevant reviews for a specific product, making the most objective matrix of reviews for it – if it has such a task. Or the most complete matrix of positive feedback, if the task is different.

The AI-assistant is a seller who can not refuse anyone's offers

Further more. When I took the next Hackathon theme, which formulates the creation of an AI-"assistant of the seller for the selection of goods in the store," even my imagination entered for a short while into a stupor. Again, the description of the hakaton did not give any information at all. And again I had to look for inspiration, zaugliv, that in this sphere others have already started to do. And somewhere among the traditional marketing bulletshit about the mob, the date, and, of course, the blockbuster, has found the following hinting text:

Cloud services of artificial intelligence and machine learning give an opportunity to interact with customers in a new way: to make interaction with them more targeted and personified. Fresh ideas of the contest participants, implemented on the basis of the intellectual cloud platform Microsoft Azure, will help M.Video on the way of digital transformation.

The tip came, which is characteristic, from one software giant: it was voiced by Tatyana Delyagina, director of marketing of data processing technology and Microsoft's cloud platform in Russia. Who as they do not know better where to dig in this topic.

If you start from her words, the AI ​​will not be used as an assistant in the catalog search, as it might seem from the category name. The AI ​​will speak to the seller, how to sell to a particular buyer. Again, letting go of the fantasy to freedom and, just in case, stipulating that the real intentions and setting of the problem of the h-katon "M-Video", which I simply do not know, it has nothing to do – with such a statement of the problem, one can again see a sea of ​​possibilities:

  1. are more or less understandable, like training AI for all the methods of a qualified salesman – from evaluating the consistency, interests, suggestibility of a person with the subsequent methods of "vparivaniya", which none of sales will never master by 100%, but the AI ​​does not have anything prevents you from working full the arsenal of all human inventions strictly by the book, giving the result of sales close to one hundred percent;
  2. and quite in the style of the "Black Mirror", when the AI ​​will analyze the background of the buyer from the transport on which he arrived and got into the cameras of the outdoor observation, to the history of publications in social networks, the history of purchases, billing and anything. In other words, the AI-assistant of the seller can be turned into an ideal sales machine, which neither spontaneous nor the most conscious buyers can counter anything.

The only hope of saving our money (perhaps by the time entering our electronic wallets under the program of unconditional basic income – well, who knows?) For an extra second between the full distribution of ideal sales machines, in this situation will be unless their ideal competition – whatever one may say , and even if the ideal AI seller Apple somehow shares my ears with the ideal AI seller Samsung – the ideal AI seller Xiaomi will have to somehow balance them all – after all, I have only two ears, the third smartphone will have to be applied somewhere in another place already.

In general, not running very much forward – I'm now very interested in what path the fantasies of the hackathon developers will go, what kind of project will the winning team have and what will they develop for the "M- Video "on. Less than two weeks are left before the results.

Leave a Comment

Your email address will not be published.