5 levels of ethical programming that determine how well smart machines will behave

People are going to have to get used to living and working alongside smart machines. With Gartner estimating that 1.1-billion connected things will be used by smart cities in 2015, there’s no escaping the fact that pretty much every object we interact with is getting smarter.

There’s a lot of hope about the future those smart objects could allow but also plenty of fear. Both Stephen Hawking and Elon Musk have, for instance, warned of the dangers of artificial intelligence.

But to a large degree making sure we don’t end up bowing down to our robot overlords could come down to how we program those smart machines.

Gartner says that realising the potential of smart machines will depend on how trusted smart machines are and how well they maintain that trust. The nucleus of establishing this trust, Gartner suggest, will be ethical values that people recognise and are comfortable with.

“Clearly, people must trust smart machines if they are to accept and use them,” said Frank Buytendijk, research vice president and analyst at Gartner. “The ability to earn trust must be part of any plan to implement artificial intelligence (AI) or smart machines, and will be an important selling point when marketing this technology. CIOs must be able to monitor smart machine technology for unintended consequences of public use and respond immediately, embracing unforeseen positive outcomes and countering undesirable ones.”

Read more: Gartner: expect mobile disruption, strong wearable growth through 2015

The technology research house has identified five levels of programming and system development, based on their ethical impact:

Level 0: Non-Ethical Programming

On this level, there are no explicit ethical considerations for the behaviour of technology. The technology manufacturer assumes very limited ethical responsibility, other than that the technology must provide the promised functions safely and reliably.

Examples include “vapourware” (a technology that is announced to the public but never delivered) that can reduce customer trust in a manufacturer. The first release of any software is seldom complete, which means customers may have limited expectations of “version 1.0” software.

Gartner recommends that technology manufacturers communicate openly on what they will deliver and any changing circumstances, altering what can be delivered and what cannot. This should include service-level agreements (SLAs) that specify what is delivered and how.

Level 1: Ethical Oversight

The next degree of sophistication has no ethical programming, but the deployment and use of technology may have ethical consequences. Smart machines may be used, but it’s essentially up to users what they do with the results. The main ethical responsibility is in the hands of those who use the smart machines.

Some companies have established an ethics board and some end-user organisations — particularly in financial services — have also established such boards but they are a small minority.

Gartner recommends that organisations establish governance practices that ensure no laws are broken, as a bare minimum. They should also seek to make ethics a part of governance by identifying and discussing dilemmas posed by using new technologies.

Level 2: Ethical Programming

This next level of complexity is now being explored, as smart machines, such as virtual personal assistants (VPAs), are being introduced. Here, the user perspective changes considerably. Whereas in Levels 0 and 1, the user is generally a business professional performing a job, in Level 2, the user is often a customer, citizen or consumer.

Responsibility is shared between the user, the service provider and the designer. Users are responsible for the content of the interactions they start (such as a search or a request), but not for the answer they receive. The designer is responsible for considering the unintended consequences of the technology’s use (within reason). The service provider has a clear responsibility to interact with users in an ethical way, while teaching the technology correctly, and monitoring its use and behaviour.

For example, one smartphone-based virtual personal assistant would in the past guide you to the nearest bridge if you told it you’d like to jump off one. Now, it is programmed to pick up on such signals and refer you to a help line. This change underlines Gartner’s recommendation to monitor technology for the unintended consequences of public use, and to respond accordingly.

Level 3: Evolutionary Ethical Programming

This level introduces ethical programming as part of a machine that learns and evolves. The more a smart machine learns, the more it departs from its original design, and the more its behaviour becomes individual. At this level the concept of the user changes again. In Level 2, the user is still in control, but in Level 3 many tasks are outsourced and performed autonomously by smart machines.

The less responsibility the user has, the more trust becomes vital to the adoption of smart machines. For example, if you don’t trust your virtual personal assistant with expense reporting, you won’t authorise it to act on your behalf. If you don’t trust your autonomous vehicle to drive through a mountain pass, you’ll take back control in that situation. Given the effect of smart technologies on society, Level 3 will require considerable new regulation and legislation.

Gartner recommends that as part of long-term planning, CIOs consider how their organisations will handle autonomous technologies acting on the owner’s behalf. For example, should a smart machine be able to access an employee’s personal or corporate credit card, or even have its own credit card?

Level 4: Machine-Developed Ethics

Gartner doesn’t expect smart machines to become self-aware within a time frame that requires planning at the moment. However, Level 4 predicts that smart machines will eventually become self-aware. They will need to be raised and taught, and their ethics will need to evolve. In Level 4, the concept of “users” has disappeared. We are left with only the concept of “actors” who initiate interactions, respond and adapt.

Ultimately, a smart machine, being self-aware and able to reflect, is responsible for its own behaviour.

“The questions that we should ask ourselves are: How will we ensure these machines stick to their responsibilities? Will we treat smart machines like pets, with owners remaining responsible? Or will we treat them like children, raising them until they are able to take responsibility for themselves?” said Mr Buytendijk.

image: Gertner