Delegate like a boss: Employ AI for your business
Tue 3 Mar 2020 | Anton Popov
What’s stopping you? Here’s a selection of intriguing and effective AI-based solutions
As of 2017, only 20 per cent of organisations were using artificial intelligence (AI) in their daily operations. While AI algorithms are still being developed and optimised, this technology already helps companies significantly improve the way they perform routine business tasks. And by allowing mundane tasks to be done more consistently and accurately, AI can provide employees with more time and mental space to perform cognitively sophisticated tasks.
Among the 20 per cent of organisations using AI, many have developed useful methods for deploying it in their operations. Here are some intriguing cases of AI deployment to encourage you to delegate your own repetitive tasks to the technology.
Many businesses have a need for video surveillance and are likely using multiple cameras over many locations. Examining the video for suspicious behaviour is not only difficult for humans, but also quite tedious. Furthermore, after many hours of staring at cameras, people are likely to miss important details.
These missteps are why AI can and should take over this task. AI surveillance tracks unusual patterns of behaviours in camera footage: for example when people stand still for a long time, move quickly or erratically, or nap in public places. If the AI detects anomalies like these, then it will alert a human attendant to the problem.
Such cameras have been installed around the perimeter of some U.S. school campuses to alert security guards to suspicious behaviour that should be investigated, for instance. To take a more unusual example, cameras equipped with motion sensors and AI are also being used to pinpoint the elephants, humans, and vehicles travelling through Africa’s vast Serengeti National Park. By alerting park rangers to behaviour that is typical of would-be poachers, it helps keep the park’s precious wildlife safe.
Whether it’s phishing, social engineering, or good old-fashioned directly stealing information via the mail, those attempting to commit fraud never cease, and those trying to stop them are in an arms race to stay ahead. Banks, credit card companies, and others have been able to use AI and its close relative, machine learning, to keep track of the typical behaviour of their account holders, using algorithms to flag any questionable withdrawals or suspicious actions.
Such methods do lead to some “false positives”, in which legitimate transactions may be blocked until additional information can be obtained. However, AI monitoring tends to become more accurate over time, as more data becomes available for it to refine its “judgments”. And such technology is definitely learning what it needs to: according to a 2019 survey of 200 financial executives, nearly 2/3 of those using AI “believe it is capable of preventing fraud before it happens”.
According to a Harvard Business Review survey, 86 percent of project managers stated they would like support from artificial intelligence in order to handle the administrative duties of a project. And tedious tasks such as finding errors or transferring information from one system to another can become so dull for human workers that they overlook data or make mistakes.
In our increasingly time-crunched society, human workers can use their limited working hours to complete more creative and high-functioning tasks. For instance, transcription services can help journalists and others produce rough drafts of interviews, saving hours of time spent typing and fact-checking. And sales departments in many large corporations have been transformed by embracing “predictive scoring”, which uses machine learning algorithms to rank sales leads by their industry, company source, past behaviour, and other factors. In this way, salespeople can learn which leads to follow up on first.
Tools like these free up employees to do what machines can only do patchily, or not at all. AI is also being used to analyse legal documents and contracts and then highlight possible discrepancies or points of contention. Similar programs have been used to help companies uncover specific information buried in an undifferentiated mass, for instance by extracting various references and mentions within millions of email messages.
Virtual assistants can be used by both employees and clients. Websites like Kono.ai and Google Calendar use AI to schedule meetings, finding the likely times in which parties are free and sometimes even handling the back-and-forth involved in confirmation.
But virtual assistants can do more than this. A retailer may use one to make predictions for what a shopper might want based on purchases that shopper has made in the past. Already in development are multichannel virtual assistants that can communicate with users on many channels, as well as virtual assistants that are capable of contextual conversations rather than just single-transaction communications.
Labelling and Identifying
Labelling and identifying tasks can be both monotonous and challenging for human workers. As an alternative, AI, deep learning and computer vision can handle multi-class labelling projects.
Take for example a recent project that the Ciklum R&D team worked on for Planet and the Brazilian government, a satellite data competition on Kaggle. To prevent the rapid deforestation of the Amazon rainforest, Ciklum’s AI team was tasked with building a model based on more than 40,000 256×256 patches from satellite images to predict 17 tags for the images. To estimate the quality of the model more than 60,000 tiles were used. The developed algorithms approached 93 per cent accuracy.
Another project was completed for SeeTree to resolve a pressing business problem: detecting when the oranges on farmers’ trees were ripe. SeeTree, an agritech company focused on tree farming, had noted that farmers don’t have enough high-quality data, and therefore had no way to start using technology to make data-driven decisions. The company approached Ciklum with a challenging task: to build an AI model that would automatically detect fruit on the trees. SeeTree provided a dataset of tree photos made with a drone. From this Ciklum research engineers built a deep learning model for automatic fruit detection that identifies fruit with nearly human precision.
Another fabulous project involved working for a European insurance alliance, which tasked Ciklum to build an AI solution to estimate the costs of repairs of damaged cars. Car claim consideration and estimating repair costs are very difficult and usually require highly skilled experts. But on their own, experts are only able to estimate to about 30 percent accuracy. With AI, the estimation process is both much smoother and more accurate.
Internet of Things
By using internet-enabled sensors that gather real-time measurements, manufacturers have a rich source of data they can use to improve their production process. By using AI to analyse this data, companies can find subtle improvements that would be nearly impossible for humans to locate on their own. Mohak Shah, the lead expert in data science for the Bosch Research and Technology Center, suggests that these technologies could reduce scraps and speed up shipping capacities.
We have only just begun to exploit the Internet of Things. One simple example of where we are today would be a thermostat that keeps the office temperature low when no one’s around, but that knows to start heating things up once a few employees have begun to filter in