Blog #2/5: AI-controlled Software and Systems in daily life


Olaf Lipinski

Having explained the basics of “Artificial Intelligence” (AI) in article #1 of this series, in this article we will take a look at where AI-controlled systems are already in use today. The (in)famous self-driving cars are not included.

In general, AI is very good at three things: Analysing large amounts of data, evaluating (also unknown) situations through statistical analysis and using the results to make predictions. Think for example of the suggestions for a next word when typing an SMS on your mobile. The next time you have a new cell phone, pay attention to how the suggestions for the next word become more accurate over time as your mobile “gets to know you” better.

It is impossible to envision product marketing without AI, where getting to know customers is key to success. On the one hand, AI “gets to know” customers personally, in terms of demographic data, buying behaviour, reaction to external influences (before the summer holidays she buys cheesy novels, in the fall text books), but also in comparison to other customers (a hundred colleagues of the same company bought books about “Job and Application” in the past few days).

In all these cases, the AI trawls through a large amount of data trying to find patterns. Every little piece of information about you is a further piece of the puzzle. The most successful seller is not necessarily the one with the best assortment of goods, but the one who always offers me exactly what I (subconsciously) want – in this very second.

Many sales are impulse-driven. The shorter the distance between the consumer and the point-of sale the more likely it is, that the sale is done. Therefore, online resellers, with Amazon leading the pack, quickly reduced the distance between customer and computer with “intelligent” loud-speakers, which also double as communication device to your personal assistant (which is AI-controlled). By the end of 2018, Amazon had already sold over 100 million “Alexas” (1) . Other successful competitors in this market are Google’s “Assistant”, Apple’s “Siri”, Microsoft’s “Cortana” and Samsung’s “Bixby” (2) .

The speed with which this entire domain is developing is evident from this article <link>, published in “Computerwoche” in 2016. The piece talks about “planned projects” (which are reality today) and that at this point in time 3 million Alexas had been sold. (In case your German is not up to speed, the AI of, for example “Google Translate”, will be happy to help: https://translate.google.com).

AI in the home

Based on the AI-functionality of “Natural Language Processing” (NLP) it is possible to talk, in normal human speech, with your personal digital assistant. Depending on the set-up, the assistant is not only able to answer questions by Internet query, but it has also access to e.g., my music library or my music streaming provider. Instead of requesting specific songs, it is possible to simply ask for “music to get into the mood for the weekend”. The AI knows what I mean.

“Smart Home” and “Internet of Things” (IoT) are currently hot topics. Not only my radiator, but also all my kitchen appliances are suddenly internet enabled. Alexa regulates room temperature on demand and is not only able to start my coffee machine when asked, but also realises, due to my known coffee consumption, that my stock of coffee will not last through the long weekend. (And thankfully automatically restocks from the online reseller!)

Today, many machines are equipped with a bundle of sensors. In a car, for example, these sensors make it possible to permanently monitor everything from current oil level to pressure and degradation of each tyre, to the current overall condition of a travelling car. The AI of the car also has access to a lot of historic data and is therefore able to quickly calculate predictions, so that the oil warning light comes on, long before imminent damage. This pro-active notification is called “Predictive Maintenance”.

“KONE”, a vendor of lifts and elevators, uses the same IoT technology in many parts of its products for “predictive maintenance” to monitor about 1 million elevators globally. The deployment of a service technician is scheduled so that s/he is on-site before parts fail and at a time the service has the least impact on the service of the elevator. In case a defect still occurs, the technician is supported by an AI to help with troubleshooting, querying databases to find reports about comparable faults and defects or helping with required documentation (3).

AI and Social Media

“Social Media” is another broad field of application for AI. The platform provider tries to get to know me as well as possible, so as to recommend relevant new friends, articles and advertising. Through analysing my profile, comments, uploaded photos, chats, videos and click behaviour, a full picture of me, the user, quickly emerges.

Next to the direct user, the “out-line” of “non-users” also emerges. If, for example, five users have Rachel Green in their address books, the AI has already the basic data of address, email, telephone number and birthday of Rachel. From event notes and GPS data it becomes clear, that these five users hang out together, regularly. The assumption therefore is, that Rachel also hangs out with the group and has attended some of these events herself. There might also be an upload of a group photo and somebody tags her image with her name…

AI also has a field day with image recognition. Thanks to good face recognition the AI sifts through all newly uploaded photos and videos for known faces, in order to tag the images, i.e., mark the images with the names of the people identified. On the other hand, it is also used to filter out all images which violate the rules of the platform, so they can be blocked.

For some time now, Twitter has been using AI search programmes quite successfully to identify accounts that are operated not by a human, but a so-called “Social-Bot”. And yes, these social-bots are also AI-controlled. They are trying to imitate humans by forwarding tweets and writing comments or replies. Their creator gave them a specific target, e.g., to depict and disseminate a certain political opinion as positive or negative. In further consideration, these social-bots are actually the perfection of “ELIZA”, the first “talking” computer application of the 1960s, which tried to imitate a psychotherapist. (see article #1 of this series) (4, 5).

Image recognition is not only used for photo and image recognition at the airport or to unlock mobiles and laptops, but also in medicine. Photos of moles can be compared against a huge database of comparable photos with attached diagnosis. The AI calculates the probability of whether the shown mole is benign or malignant.

Another field of application for image recognition was identified by oil producer ADNOC (Abu Dhabi National Oil Company). ADNOC visually examines and classifies drill core samples. Previously, this job could only be done by very experienced geologists and only a limited number of cores could be examined per day. Experts are a scarce commodity, and when one retires, his experience also retires. By using AI, the number of cores processed daily could be increased massively (527 images per second). In addition, younger geologists can now be deployed and gain experience in this field (6).

AI in service centres

To close this article, I would like to point out an area of application with which most of us have already had contact: Chat-bots and AI-controlled software in service centres. You know chat-bots. They are the friendly colleagues which open on every other website in the bottom right corner and desperately want to help you. The functionally of a single chat-bot is normally limited. It cannot do a lot of things, but what is does, it does well. With a list of 5 – 10 greetings the bot knows what a greeting is and recognises one, even if it has never heard this word before (AI). For the other blocks of a conversation, the same principle applies.

It is fairly easy to give this chat-bot an AI-generated voice, which does not sound at all tinny anymore. And presto, the bot is able to telephone with those customers who only want to enquire about opening hours, the web address or any other standard information. Ethically correct chat-bots will advise at the beginning of the “conversation” that they are not human, but are all companies doing so? In any case, these bots are helping human service agents by taking care of time-consuming standard questions and freeing humans for questions requiring expertise and judgement. In addition, a chat-bot does not need to be home in time for dinner, but does its shift 24/7.

Thanks to AI-speech recognition, a customer is able to explain in natural language what her concern is and is then transferred to the correct line. An end to “Press 1 for … 8, in case you want to hear it all-over again” [tinny voice]. Against this backdrop you start to realise why calls with service centres are recorded for training purposes. It is not necessarily a human who is being trained, but maybe the chat-bot.

In order to increase the speed and quality of human agents, administration of documents and tickets is also being done increasingly by AI. In case of a contextual query, the likelihood that useful information is retrieved, is far larger, if it is done by AI, than for example with a keyword system.

Due to the fact that many customers are already used to communicating with Siri, Alexa and Co. for them chat-bot is the first choice to get in touch with a company – and this behavioural trend is the same across generations (7) .

This closes the circle. As you can see, AI-controlled software has been around for a while in our daily lives – and that was only a small sample of applications.

The next article of this series will have a detailed look at how AI is used for test automation.

By the way, in case you are not sure whether it is a human or a chat-bot at the other end of the telephone line: Just ask exactly the same question a couple of times in a row. The chat-bot will answer identically every time.

This article is a collaboration of: Olaf Lipinski, Wilhelm Kapp and Dejan Husrefovic, SwissQ Consulting August 2020

 

Sources

(1) https://www.handelsblatt.com/unternehmen/it-medien/sprachassistent-amazon-hat-mehr-als-100-millionen-alexa-geraete-verkauft/23830954.html?ticket=ST-1488897-OStfLo6tkS1J106E2Gy9-ap5, accessed 07.08.2020
(2) https://www.homeandsmart.de/smart-home-sprachassistenten, accessed 07.08.2020
(3) https://www.ibm.com/watson/stories/kone/, accessed 07.08.2020
(4) https://www.handelsblatt.com/technik/it-internet/kurznachrichtendienst-im-kampf-gegen-fake-news-holt-sich-twitter-ceo-dorsey-hilfe-aus-den-niederlanden/22862266.html, accessed 07.08.2020
(5) https://www.fr.de/kultur/twitter-loescht-millionen-fake-profile-10964716.html, accessed 07.08.2020
(6) https://www.ibm.com/case-studies/abu-dhabi-national-oil-company-adnoc, accessed 07.08.2020
(7) https://outgrow.co/blog/vital-chatbot-statistics, accessed 07.08.2020

0 thoughts on “Blog #2/5: AI-controlled Software and Systems in daily life”


Leave a Reply