Category: ARTIFICIAL

We have recently accepted the idea of omnipresent AI adoption, but new changes and technology reshaping are already on the doorstep. Have you heard of the Responsible AI concept? Let’s take a closer look at it, since it will soon become a mandatory tool to build business relationships.

Article by Dmitry Doshaniy, NNTC General Manager
LinkedIn | Mail

Proofs of Concept are in trend now

Enterprises usually struggle to develop AI pilots into production. Unfortunately, many solutions, when facing approval and development obstacles, never overgrow the “Proof of Concept” stage. You might encounter such a situation too. Being concerned about disruptive effect of innovations, most businesses just try them out in a few cases and do not go further. Thus, robots are trusted to make the first lead touch only, with all other steps being usually assigned to a human being. This is why such robots are not expected to do more than read, write, speak, and copy/paste.

In such circumstances, the value and business potential of AI cannot be unveiled and win the so much needed customer confidence. Meanwhile, AI-based solutions can completely change the approach to working, training employees, ensuring people safety, as well as dramatically speed up business processes. So where is the problem?

AI communication problems

The first communication problem arises between participants of the solution implementation process. Unfortunately, product developers and customers speak different languages and often focus on different features of the proposed solution. Due to such misaligned visions of AI and its potential, the technology capabilities are underutilized.

The second communication problem arises when a customer interacts with the final product and perceives it as something hostile or suspicious due to a lack of information about its benefits and functions. Let’s consider recruiting robots. HR people are more concerned about their jobs and afraid of robotic automation possibly causing lockouts rather than interested in studying the solution capabilities.

How does the AI Responsible approach solve them?

It is simple. This approach assumes transparent unified communication with many target groups, and can be used by both AI-powered product developers and employees who are responsible for implementing artificial intelligence solutions in their companies.

AI-based decisions need to be explainable in order to be trusted. According to a recent PwC survey, 84% of CEOs agree with this statement. As well as buyers. Consumers want the convenience of services tailored to their needs, together with the peace of mind knowing that companies are not unknowingly biased against them — and that their government will protect them with laws regulating how their data can be used. As we can see, being just useful is not enough. It’s important to get clear, and we at NNTC go this way with each of our products.

The “AI Responsible” formula has five key components to identify AI responsibility:

• GOVERNANCE will introduce clarity to AI objectives and tasks within an organization. Before developing or implementing a solution, assign and document the responsibility of each process participant. Then, align AI with the current business strategy and identify processes that need to be optimized using AI. What result should we expect? How will it manifest? Who will monitor the effectiveness and document problem areas? Can you give guarantees that the solution will be effective?
• ETHICS AND REGULATION. AI-powered solutions must be developed in accordance with the established legislation and regulations of the organization, which should be morally responsible and ethically defensible. Notifying of compliance with the established guidelines will build confidence in AI.
• INTERPRETABILITY AND EXPLAINABILITY. Let the user understand why a particular AI reached a particular decision. Explanations should be tailored to the different stakeholders, including regulators, data scientists, business sponsors, and end consumers: never leave them uninformed making your product into a mysterious black box for them.
• ROBUSTNESS AND SECURITY. Eliminate all possible risks and flaws in your product, and timely release necessary updates and patches. Tell your customers what problems have been solved, quickly fix bugs, and listen to feedback.
• BIAS AND FAIRNESS. Fight against discrimination and train your AI to perform its functions fairly and equitably.

Only those who are responsible will survive

The ability to develop AI responsibility is a cornerstone of its further development and recognition. In a world filled with a variety of solutions from styling photos like watercolor paintings to recognizing car license plates, the advantage will belong to a reasonable developer who explains algorithms and is ready for a dialog with users.

Read more

• How to win and retain customers after the Pandemic?

A mess on a construction site is quite a common thing and can be tolerated to some extent if it is an organized mess. However, when this mess creeps into core processes, a construction site turns into a hurrah’s nest, and a developer faces risk of large penalties, downtime, and even death toll. In this blog, we’ll tell you how CROC, our strategic partner, implemented a face recognition system on a construction site.

One fine, sunny day, when the site was buzzing with activity and dozens of workers were moving around cheerful and positive, a former workman penetrated the site. He passed through a checkpoint, greeted a guard who knew him well (but did not know that he had been fired already), entered the territory, and went to his friends. Half an hour later, a fire was burning, with the workers drinking tea, resting, and enjoying their lives.

The problem was that they made the fire near two gas cylinders, one of which suddenly exploded. Neither warning nor alert. Just exploded, and that’s it! Fortunately, people were just shocked, not injured, but a police came to the site the same day and initiated an inspection, which revealed some unpleasant facts for a contractor:

• Many workers used forged documents. In other words, one family member obtained necessary documents, got hired at the construction site, and then his brother went for work instead of him.
• For each worker using forged documents, a fine of $7,600 or even more may be imposed, and the construction process may be suspended for 14 – 90 days with almost catastrophic consequences for a general contractor and customer.

That is why the contractor asked CROC for help.

How to solve a problem?

Initially, the construction site was equipped with just run-of-the-mill employee recognition systems:

• Palm vein identification, which performs poorly even in clean offices if people eat some greasy food there. After eating, this vein pattern changes causing recognition errors.
• Iris identification, which is more expensive and time consuming. It is hardly the best option when every morning buses arrive at a construction site and some 600 workers have to pass the checkpoint within one hour, from 7 to 8 a.m. It’s as slow as a snail!

Therefore, it was the time to try video analytics.

Each turnstile is equipped with two cameras: one aimed at incoming people, and the other, at outgoing.

There is a small cooled server room on site, having a rack with two servers and a switch. The first server receives video stream from cameras and identifies frames containing faces and being of the best quality. The second server receives these “clear” frames with faces and runs a recognition algorithm to find a match in a database.

If a frame contains only one face and this person is authorized to pass through, then the turnstile opens. Otherwise, a guard is alerted.

No Internet connection is needed for the system, just a copper cabling on site. However, if connected, the system allows for continuous monitoring, email reports, and remote second line support for operators.

The system runs perfectly in typical situations:

• Using a pass card of another person to enter — an alerted guard stops a trespasser and investigates the incident.
• A former employee attempts to enter the site — a guard does not let him in.
• An employee attempts to enter the site in wrong time — a guard asks the employee to get a pass card at an access control office and then come back to the checkpoint.

In addition, every evening, executives get a report on how many people were scheduled for work and how many arrived actually. It is very important since contractor’s favorite trick is to schedule 50 workers, provide only 30, and then report on 50.

Moreover, this information — like 600 people scheduled, only 450 arrived — shows if a project is running out of time already. And you see it right now, but not a week or month later.

Outcome

Morning queues have almost disappeared. Instead, it takes one hour for two turnstiles to let 700 people in, even including guard intervention cases. The customer enjoys reports on the actual number of people arrived at the construction site, which was a sad news for some of the contractors. Neither a $7,600 penalty nor work suspension for 14 – 90 days is pending now, with the system itself costing less than this penalty.

NNTC actively partners with CROC and implements face recognition solutions in GCC countries. If you are interested in this technology, feel free to contact our consultants.

Robotic Process Automation (RPA) is about optimizing individual processes and freeing up people for human-only tasks. Robots will never completely replace humans, that’s for sure, since feelings, emotions, and critical thinking are all beyond machines’ capability. However, robots excel at calculations and simple tasks – no hard choices or emotional investing.

According to PwC Middle East research, the AI market in the Middle East will rise to $320 billion by 2030. There is a strong focus on RPA of customer care services: in 2017, the Dubai Electricity and Water Authority (DEWA) announced recruitment of five robots to its Customer Happiness Centres. Retailers also tend to engage robots. Pepper, a world famous robot, welcomes visitors at several UAE banks and salespoints, defines customer’s gender and provides recommendations for a retail demo or special offers in the store.

For robots, the retail sector is an ideal testing field and promising application area. Just look at numerous use cases, like automated store shelve control (merchandising, stock and price tag checks) using robotic carts with optical recognition module – a godsend for merchandisers, or cashier workload calculation based on video analytics to better schedule salesforce working time in a shopping space. And what about PoS self-service payment system powered by face recognition? Such a fare system was rolled out in Chinese subway and Amazon Go self-service stores so we can say for sure that it really works. 

In warehouses, robots handle and sort products. Besides other things, Amazon is famous for its robotized warehouses, promptly processing millions of orders every day. Dozens of robotized auto-loaders, as little as a common vacuum cleaner, easily move heavy shelve stands with goods to operators who just take necessary items from the shelves. As a result, with a help of a robot, one operator does the job of 6-7 persons, with order batching becoming cheaper as well.  With the routes determined in advance, robots avoid collisions and easily transport big shelve stands within a limited territory. Adler, a German clothing chain, leverages robots for warehouse inventory control. Every night, robots called Tory scan RFID tags on goods and generate stock reports. 

However, robots cannot replace all sales personnel when it comes to a bit more sophisticated tasks, as we can see from Amazon’s experiment with robotized stores. The company tested its first Amazon Go stores back in 2016, with 10 such stores currently operating in the U.S. No cash desks, cashiers, consultants or queues there. At the entrance, shoppers scan their Amazon Accounts and at once find themselves among smart sensors.  The system tracks what goods shoppers take from shelves and how they move across the store, and automatically debits the necessary sum from the buyer’s account once they exit the supermarket. Despite the marvelous idea and company’s attempts to improve the system, the stores regularly face difficulties as tracking more than 20 people moving across the store turned out to be a real challenge for the cameras. If a shopper takes an item, reads a tag and then puts the item back, the system can lose sight of this item because of a changed location. In addition, you’ll hardly find alcohol in such a store because alcohol sales require shopper age identification. 

AI improvement and education efforts may take a long time before innovators can introduce new generation stores operating without cashiers or other personnel. To address this challenge to a certain degree, we have developed an unusual motion detection (UMD) solution to identify shoppers’ atypical behavior. Here, the first implementation stage is all about UMD solution continuous learning based on real-life customer data, taking up to two weeks. The system collects data from video cameras and, after learning, can quickly detect atypical behaviors: a person is running instead of walking, stops where shoppers usually do not stop, zips a coat too slowly, carries a big backpack in his hand, or wears baggy clothes. 

Another challenge for RPA adoption in the retail sector relates to vulnerabilities and hacking attempts – something both competitors and tech-savvy cybercriminals may exploit or opt for. Moreover, any Internet-connected system is exposed to DDoS attacks. Yes, it is. And robots as any artificial Intelligence tools, are exposed too. Modern RPA systems employ advanced security technologies resulting from an ever-lasting competition between cybercriminals and corporate white hats. Furthermore, when it comes to software robots, they are made fault-tolerant and definitely have proper and reliable protection, thus being a good choice for any retailer. Opportunities always go hand in hand with risks and it is an innovator’s responsibility to prudently assess such risk-to-opportunity balance and adopt the best possible approach using verified data and battle-proven technology.

In this infographic you will find the best statistics about artificial intelligence in the Middle East and find some quality insights for your business.