Helping Democratize AI, A Call to Action

Original article was published on Artificial Intelligence on Medium

Helping Democratize AI, A Call to Action

A call to action to help democratize AI. To participate please visit Surveyhero

Source: Google Images

Companies across the globe are increasingly adopting Artificial Intelligence (AI). In Gartner’s AI-adoption-in-the-enterprise-2020-report, only 15% of surveyed respondents have stated to neither be working with nor evaluating the deployment of AI.

Meanwhile, forecasts for its global market value are skyrocketing. AI has the potential to save businesses substantial costs and can generate new revenue in ways that wouldn’t have been possible before.

A study by Accenture and Frontier Economics found that in manufacturing alone, AI could lead to a worldwide output boost of 4 trillion US$ over the next 15 years.

But there are different hurdles to cross to get a piece of the pie, and not everyone fulfills the necessary requirements to benefit from applications of artificial intelligence.

Companies of different sizes and origins are affected differently by these hurdles. Especially big corporations who have enough money to pay the high wages data scientists currently receive and generate sufficient internal data for the training of accurate models. Startups also have an advantage in the field as they are established on the new fundamentals of technology and typically are a lot more agile and tech-affine than more mature companies. Most disadvantaged are established small and medium-sized enterprises (SMEs) that are still occupied with the digital transformation of their businesses. As SMEs “represent about 90% of businesses and more than 50% of employment worldwide” (Worldbank), they should not be neglected!

The AI-adoption-in-the-enterprise-2020-report found that the top five leading bottlenecks to AI adoption are:

1. nonrecognition of need,

2. difficulties in the identification of appropriate use cases,

3. challenges in hiring the required staff,

4. lack of data quantity or quality, and

5. challenges of technical infrastructure

Let’s start by examining the first two problems. Opposed to the last three, they have instead focused on the culture around and experience with AI than on the resources required for implementation.

Gartner’s research shows that nonrecognition is much less of a problem in companies who are already deploying AI in a mature state than in those who are still evaluating it.

Regarding the difficulty of identifying appropriate use cases, one might say that this is a childhood disease of any technology when it hits the market.

When it comes to the required resources for the implementation of AI, let me quote Dave Chapelle:

Source Meming World

The last three of the five issues are tackled by what can be referred to as the ‘Democratization of AI.’ As the expression suggests, it is about enabling a broad spectrum of the economy as well as society, in general, to benefit through the advantages of AI. Significant examples are:

Automated Machine Learning

A variety of providers, ranging from tech giants like Google or Amazon to university teams like Freiburg-Hanover, offer so-called AutoML solutions that aim at automating the Machine Learning pipeline of model creation.

Cloud-Computing and Storage

Deep Learning, a subset of Machine Learning, requires significant storage for training data and specialized processing clusters to enable efficient training of models. To invest in proprietary servers and GPUs or TPUs to boost memory and accelerate the processing power, before having identified a feasible use case can be a big turn-off for the procurement of any company. But there are options on how to surpass a large fixed cost investment upfront: simply rent the required technical infrastructure in the cloud. Just like with AutoML, there are dozens of providers from startups to large multinational corporations.

Pretrained models and Machine Learning data repositories

Furthermore, there is the possibility to use pre-trained models or training models on data that finds its origin outside of one’s company. There are many ways to get access to data on the internet. Options like Kaggle, the UCI Machine Learning repository, or OpenML already enable the training of traditional Machine Learning and Deep Learning models on a broad scope of different data from various application fields.

This is where I want to extend the current state of research. Insufficient data, be it for reasons of quality or quantity, is a problem in the adoption of AI not only of SMEs but for companies of all sizes. Together with the help of those who are already training their models fully or partially on external data, the focus of my thesis is the revelation of data selection criteria, data providers, types, and formats across countries, industries, and organizational units.

If you want to help push the boundaries of AI democratization, I would be very thankful for your support in gaining insights on external data in the context of Machine Learning. Participation takes up to 10 minutes and is anonymous. The outcome will be shared afterward so as many people as possible can profit from the results, which are the centerpiece to my empirical study.

✅ Please participate here: Surveyhero

A little side-info about me: I am a student originally from the south of Germany living in our capital since 2018. I have graduated with a Bachelor of Science in Business and Economics from Goethe University in Frankfurt. I am currently finishing my Master’s degree in International Marketing Management at the Berlin School of Economics and Law. For my thesis, I have decided to focus on Artificial Intelligence, as I believe that there is a high potential that it will automate my field of business. I am strongly interested in the possibilities our future bears for us. I chose the topic of democratization because I think it serves a great purpose in the means of creating equal opportunities in a field that bears the risk of extending the wealth gap across the world. If you have any questions, feel free to reach out to me.