Can it be trusted?

AI's big challenge

Think artificial intelligence is some far off technology on the distant horizon? Not so. It's not only here, it's all around you. Our Torsten Mallée examines a major hurdle ahead.

AI and the transparency challenge

Artificial Intelligence seems to be one of the most discussed topics in business these days. And, as happens so often with hot topics, opinions are extreme.

Partially, this is because AI is kind of a blurry umbrella term for what computing power and data paired with algorithms can do.  In fact, the discussions span over AI as a supporting technology for business, as a replacement of human workforce, up to visions of AI dominating mankind.

So, has AI become just another over-hyped topic? No, not in my opinion. 

AI is reality already today and is in many areas of our daily lives. There are smart home speakers like Alexa or Google Home. Visual search apps that find offers for products that you have just simply taken a photo of and there’s speech recognition or text recognition for your smartphone. These use algorithms, like those from Netflix or Spotify, to make proposals for movies or music that match your personal preferences. 

Artificial Intelligence
Artificial Intelligence

I, myself, love these functionalities as they make life easier and are simply just cool. I like searching for photos on my smartphone by using terms (try typing “drink”). Actually, I cannot wait to see these helpers get even better.

Why AI deserves the attention

While the presence of AI in our lives is not really noticed, what is it that fuels the interest and investments in(to) AI?

Just as described above, AI needs mainly three things: high computing power, high amounts of data and intelligent algorithms.

And, just these ingredients required, explain why the relevance of AI has grown so fast in the recent past. Algorithms, albeit not easy to program, are more or less simply a matter of brain work and continuous improvement.  Computing power and data were their limit. This has changed drastically.

Artificial intelligence
Artificial intelligence

Today, computing power has achieved levels that allow AI to be really sophisticated. For example, take the A12 bionic chip in Apple’s latest iPhone XS. It is able to process 5 trillion operations per second. If this is possible in a smartphone already, it is easy to imagine what big machines can do.

Data is ubiquitous in our more and more digitized world. With sensors becoming less and less expensive and software that controls more and more processes, the remaining challenge is the quality of data. In many areas, this must however not be underestimated.

So, are we all ready to go and can we just start deploying AI in our businesses? Well, it is not so simple. There are many challenges and typical mistakes to avoid in respective projects. A quite comprehensive summary has been compiled by Google’s Chief Decision Intelligence Engineer, Cassie Kozyrkov, in her article "The ultimate guide to starting AI“.

Artificial intelligence
Artificial intelligence

The trust challenge

Apart from the more project-related challenges to successfully introduce AI in your business, there is however another important topic to address. This is that individuals just do not trust AI when it is about important decisions. 

And, here is the important difference between the private use - where AI is just a nice helper - and the professional use where AI has a bigger impact. This  lack of trust creates quite some resistance I've noticed when discussing this topic. The reasons mentioned for doubts most often relate to potential faults in algorithms used or the data that these algorithms work on – or both. It is within human nature that trust needs be earned first.

And, I do agree. Quite a portion of this skepticism is well justified. First, we all agree that the sheer existence of data as input for AI does not mean, that it is the right one in the right quality. Second, it is comparably easy to come up with algorithms that prove to be right based on past data. However, this does not guarantee, that these algorithms draw similarly correct conclusions based on future data. These obvious reasons given, businesses should be skeptic until proven otherwise – especially, when it is AI-based critical decisions that touch risk and compliance matters.


AI must be transparent

Exactly this challenge for AI requires to establish a sober view on what AI can and should look like when starting to use it: AI must be transparent! Users do want to understand not only the “what” but also the “why”!

But, wait, can AI at all be transparent? Eventually, we talk about the combination of immense computing power with huge amounts of data. How can humans comprehend AI-based decision making then?

Well, it is possible. To achieve a satisfactory level of transparency – which I would define as the same level of transparency that humans have for information they use as a basis for their decisions – it, in my opinion, is sufficient to see the aggregated findings of AI that it finally draws its decisions from.


In an example in the world of supply chain management, i.e. for inventory management, this could look as follows: AI suggests stocking up inventory for a silver BBQ-grill beyond usual amounts in a certain region. The manager would now be able to see the main criteria this proposal is based on in some dashboard view: an aggregated summary of statements from social media indicating extraordinary buzz around the product, weather forecasts, sales figures of last year, a preference for the product in silver color in the target region, statistical lead-times for the product etc. Such information would be good to convince a manager, that the suggestion is based on valid reasons.

As the example shows, transparency comes at a cost, though. If AI-decisions shall be transparent, meaning traceable by humans, this will restrict the number of parameters and the decision logic complexity. So, non-transparent (opaque) AI systems are potentially more powerful than transparent AI-systems as they have no restrictions concerning their input data and reasoning. In a few cases, this might be a trade-in for achieving higher acceptance among users.


Transparency is the entry into successful usage of AI for critical decisions. For the time being, AI is used best to complement human expertise where it can come up with insights that humans cannot create at same cost and quality.


Maybe someday, nobody will question AI for certain use-cases anymore and the opaque AI is accepted,  just as you do not question decisions of employees or colleagues (are they fully transparent to you?). 

So, AI eventually is not about a replacement of humans. This is a challenge that is still to be managed. It is about enhancing and speeding up human decisions.