When is AI the best solution?

Any organisation considering the purchase of an AI system needs to determine

• what problem or optimisation the AI is intended to solve
• that AI technology is the best solution for the objective
• how the system will be configured to serve the chosen objective
• the rules and boundaries of use of the AI system.

In the public sector, the primary motive must always be a better service and easier everyday life for citizens, communities and businesses. Making administration more efficient and saving costs will always have to come second.

Consider:

1. what kind of change your organisation is aiming for
2. how the future operating environment will differ from the current one
3. why you would choose AI technology of all things as the solution
4. how the procurement implements the fundamental rights enshrined in the Constitution
5. how the procurement implements authorities’ obligation to promote equality.

There is no point in turning all digital systems and applications into autonomous algorithmic systems. An AI solution makes no sense if the desired benefit and efficiency can be achieved with a simpler solution, such as rules-based automation.

You have to understand the nature, advantages and constraints of different solutions and, after objective consideration, choose the method that is best suited for the pursued objective. To do so, you need to collaborate with experts in technology, substance and business.

First and foremost, AI systems are basically efficient statistical and probability computing machines. Their superpowers are

• fast processing of big data
• ability to process different datasets from multiple sources simultaneously
• ability to look at large datasets and detect connections and regularities that are difficult for humans to notice
• ability to produce forecasts, analyses and summaries of data.

AI can be used to introduce efficiency and new insights to tasks that would be difficult or even impossible to manage with human effort and traditional technology alone.

Areas of application include

• improving the efficiency of production processes
• automation of repeated and heavy processes
• mapping and anticipating customer needs, including accessibility and multilingualism
• preventive health care and early diagnosis of diseases
• proactive community planning and environmental planning
• energy efficiency and clean energy development
• mitigation of climate change and biodiversity loss
• detection of crime that is planned and carried out online
• cybersecurity
• prevention of tax evasion and money laundering.

That is to say, the potential benefits are very significant for people, the economy and the environment.

AI systems are not magic or anything new. They are advanced computing systems directly developed from preceding information technology and data science. They are machines that abide by the same rule of thumb as with all information technology: garbage in, garbage out. Bad data can only produce bad outputs, even if your intentions are good.

Machines do not know what is essential or true. And they don’t care either way. AI systems process biased, outdated, even discriminatory or completely fabricated data just as well as relevant and factual data that promotes equality. The data just needs to be in the correct format for processing.

Even advanced AI systems require human-defined criteria that can be used to assess the quality and appropriateness of the output. These criteria cannot be defined, at least responsibly, by anyone else than those who are deciding on the objectives of the system.

When a system works in a different way than intended, the reason is almost always discovered in the actions of the people responsible for the system. Research has found that it is surprisingly difficult to define accurately enough what an AI system is expected to produce.

The more elements and changes there are in the operating environment of an AI system, the more difficult it is to define a precise objective for it. Oftentimes, administrators also have several simultaneous objectives for their systems, some of which are hidden as secondary objectives of sorts.

For a rule of thumb for accountable development, we could turn to the classic “small is beautiful”. When your organisation focuses on a limited and clear-cut issue or task, it will have an easier time formulating the task as an objective for the machine and achieve usable results.

Success is also essentially influenced by your organisation’s ability to identify relevant issues and possible solutions to them.