Do hidden robots need guiding standards too?

Even back in 1942, there were dreamers about what the days of artificial intelligence would look like. Futurists like Isaac Asimov were considering the risks of new autonomous technologies. It was during that year that Asimov wrote a short story entitled “Runaround” in which he unveiled the three laws of robotics.

The key theme for these laws was that a robot could not through action or inaction allow harm to come to humans. Over the years both philosophers and writers have examined these laws in myriad ways showing the loopholes in the language and the challenges that can arise in edge cases. Regardless, the principles seem like the sort of thing we’d want if robots walked among us. They should serve to enhance our lives.

If you’ve ever seen a video of Boston Dynamics’ robots, you understand why the three laws are needed, at least at an emotional level. Boston Dynamics makes all sorts of animal/human-like machines and they seem like something out of a science fiction movie where the robots are not benevolent servants but instead determined to be our overlords. The videos of those robots are evidence to support the need to get those laws right before Atlas walks among us.

But what about the hidden robots, the robots that exist only as lines of code buried on a web server in a cloud hosting facility and don’t look menacing? Should we also be giving thought to guiding principles of design for these engines that are fed our data and are allegedly supposed to make our user experience better?

It seems like a no-brainer. However, anyone can sign-up for their own cloud-based hosting account which likely includes a machine learning starter kit. With a little skill and the right data, a journeyman data scientist can create technology that can do things that would have seemed magical twenty years ago. In the hands of more talented operator far more extraordinary possibilities exist. So what responsibility do each of these developers have to society before they unleash their machines upon us?

I suspect that the European Union is going to lead in this space much as they did with privacy. I also suspect that the initial laws of robotics/AI are going to me more focused on disclosure than compliance with behavioral norms. But this is the sort of thing that could get out of hand, not in the Skynet manner but more in the way that Facebook struggled with privacy. The technology will be two steps ahead of our understanding of how both it, and the humans who created it, will be using it.

I’m optimistic about the possibilities for AI to have an almost magical ability to improve many aspects our lives. But like with privacy, I think we have to be looking forward to the risks that such technology to have a negative impact. We need to be intentional about ensuring that the machines are learning to work to our benefit.

Steve Zakur

About Steve Zakur

Stephen Zakur is CEO of SoloSegment. SoloSegment provides analytics that improve site search conversion and machine learning technologies that improve content effectiveness.

Clarke had it right, AI is magic

Any sufficiently advanced technology is indistinguishable from magic


Arthur C Clarke

It seems like AI has been on everyone’s minds lately. It definitely has been on ours, as Tim Peter and I spoke on AI on our latest podcast. AI has been particularly hyped up, with plenty of big ideas emerging about what it can do for website owners. But I’m fearing, that like blockchain, we’re heading for Gartner’s fabled Trough of Disillusionment if we’re not there already. AI can’t solve all your business problems, though there are those that are well suited with the tools that are available today. But like any solution you have to have a valuable problem and the right approach to applying the solution.

So, how do you get started? There are three real impediments to getting AI off the ground.

  1. Unreasonable expectations
  2. Concerns about data
  3. Skills and Experience

The AI Expectation Problem

We always overestimate the change that will occur in the next two years and underestimate the change that will occur in the next ten. Don’t let yourself be lulled into inaction.



Bill Gates

The Trough of Disillusionment is largely filled with folks, especially at B2B companies, who came to AI with unreasonable expectations. Like any new technology our expectations for near-term impact are always too high. There are no magical powers, there’s only hard work. So the first step in applying AI to any business problem is assessing the measurable value of the problem (make sure you have a business case) and think small.

Most “big bang” projects — large budgets, lengthy schedules, massive business cases — fail to meet expectations. With new technology the risk is even greater because not only are you proving that the project is valuable, but also that the platform can deliver.

To minimize your risk, think MVP (Minimum Viable Product) which is really just a fancy way of saying “Proof of Concept”. Identify a handful of experiments that you can run. This reduces the risk of failure — the likelihood that all the experiments fail is low — and set out goals that aren’t purely business value. For instance, teaching your dev team how to set-up a text analytics platform has a lot of value in the long run.

The AI Data Challenge

One of the intimidating challenges for AI projects is getting the data. Modeling can consume a fair amount of data but it’s not usually the volume of data that trips companies up, it’s that availability of that data. 

Many problems where AI can help requires data from across the organization. Building the connections, both technically and within the management system, with other organizations to access the data is critically important. Ideally, availing yourself of data from work that’s already being done within the company will provide you with the right access. Of course, normalizing that data to work together can still be a challenge.

The AI Barrier: Cost

One of the largest barriers to getting started is skills and expertise. Competition for data scientists is fierce and consultants who do this work can be costly. There are essentially two types of consultants that can help. Domain experts with software that focuses on one specific type of problem and custom development shops. 

Working with a software vendors can provide you with a quick start, but it often presumes that you have a problem that fits with the software that they’re selling. What we’ve seen in the marketplace is that the best packaged AI solutions are in very narrow domains. If that’s a fit for you it can be a great accelerator.

Custom development is a great option when you have a rather unique problem. The downside of this approach is that you’re often building both the platform for the application and the application itself. The timelines for this approach can be long and the cost high. 

One of the the ways we’ve found successful is to find a vendor who has both domain expertise and a good platform but not necessarily an application that meets the need. If they have application expertise in a close swimlane, they may be able to provide you with something that is specialized for your use case but not rigid like a prebuilt application. This allows you to enter with a modest investment and a solution that meets your solution needs.

It’s not magic, it’s work. Valuable Work.

When AI works, I think Clarke was right, it does seem magical. And what business can’t use a little magic? But don’t buy into the hype. Don’t be frightened by the expectations curve. Do find a valuable problem. Do run a few experiments. Do start. Build the muscle memory. Find the place where AI allows you to build a valuable customer experience.

Originally posted on Biznology

Steve Zakur

About Steve Zakur

Stephen Zakur is CEO of SoloSegment. SoloSegment provides analytics that improve site search conversion and machine learning technologies that improve content effectiveness.