These days AI is on everyone’s lips – and it’s one of the strongest powers for change in IT. An area in AI that’s growing fast is digital assistance. It’s a computer programme which, with help from artificial intelligence, performs advanced automised interactions that can receive and manage orders or offer more effective management of systems and equipment through speech and text. But how do you hire a digital assistant in a safe and cost-effective way?
To begin, we need to understand the system’s prerequisites, strengths and weaknesses.
Modern AI systems are excellent at handling large amounts of data or seeing patterns, and can be incredibly fast in decision making when the system is trained correctly. On the other hand, its abilities within areas such as creativity, fantasy and social interaction are far behind human ability.
AI systems usually consist of a number of layers that extract information and help make decisions. These layers need to be carefully trained and for this, large amounts of training material are required – such as images or text, fed through the system and guided during training to teach the system what is ‘right’ and ‘wrong’.
Because the education of the system needs to be under a controlled environment, most of today’s AI systems have their learning ability switched off. The system is first trained, and then put into production in a ‘frozen mode’. This is important to know when installing an AI system. After all, you don’t want your digital assistant in customer services to become rude and unpleasant if it learns from customers how to be rude and unpleasant. A well-known example is a Microsoft chatbot who, through communicating over Twitter, developed racist behaviour.
Fortunately, many AI systems already have capability for linguistics and language. Upon introduction, focus can be placed on operational capabilities, for example: performing specific tasks within a business or organisation. Interactions can be entered into the system’s regular engine, or programmed specifically for a purpose. If these interactions are so complicated that they cannot afford to be overlooked, robot training should be considered. Support systems for military or medical care are good examples – there can be multiple parameters to take into account and a large amount of information is used as a basis for decision making and therefore, training is a must.
Another area that is important to take account of is the law or responsible body that the digital assistant uses. Do you give the digital assistant the ability to give advice or make decisions? Who is responsible for these decisions or advice, and how are they handled legally?
This is a largely untested area and many users of AI systems choose to use a system that only functions supported and supervised by real people. This way, decisions can be made quickly and correctly with the help of AI, with the responsibilities and legal aspects working exactly as before.
So if you want to avoid legal problems you need to be careful with what the digital assistant is allowed to help with. But on the other hand, there are large financial gains to be made through shifting the boundaries of what digital assistance is used for – provided there are the resources for it.
The third area that’s important to manage is integration. For a digital assistant to deliver real value, it’s important that they integrate with existing operating systems. Information needs to be collected and sent back like an order.
For integration work to function well, it takes planning. The information used by the digital assistant must be easily accessible and the AI system needs good integration support. Most importantly, the AI system and operating systems should use the same integration technologies.
The introduction of a digital assistant does not need to be too complicated. An introduction can be started using a relatively simple digital assistant that only provides a limited controlled functionality – this is how most digital assistants are used today. The difficult and expensive training can initially be ignored. Instead, the focus should be on supporting business processes and ensuring that the improvements you hope to achieve with the digital assistant become a reality.
Johannes Jansson
Consultant & Associated Partner
