fbpx

Why should we make a more vigorous effort to democratize AI?

Democratize AI

Let’s face this problem: Implicit bias means everybody has their own bias according to their livelihood and life experiences. The consequences of implicit bias in AI are catastrophic. Since most of the time, AI systems are designed by relatively few private organizations and corporations worldwide. They include their own bias and interests into products and services impacting millions of people’s lives—the foresight solution: Democratizing AI.

Advertisements

Just look at the AI Patents institutions in 2020. They are clearly concentrated in the USA big tech companies. This is a indicator of how these corporations develop their own AI capabilities for their business objectives.

AI Patents 2020
IFI CLAIMS on AI patents

One of the key issues that provoke this power’s concentration is the lack of government regulation over the data used to train the models. Since these large corporations sit in a colossal amount of data, they do not hesitate to monetize that data. Regulators barely understand the value of these organizations’ knowledge, which seems too complicated to set rules for that information’s right usage.

Some eye-popping troubles of not getting rid of the access’ barriers in AI are:

  • The fear of AI slows down the adoption of AI solutions which, in general terms, unlocks the improvements in the productivity of companies and society, translated to a lack of advancements in people’s well-being.
  • Without research funds, there are no chances to have a world-class science, thus not creating innovative ideas.
Advertisements

To democratize AI we need to explore the barriers of AI general understanding and adoption:

Myths to overcome about AI

  • You need a computer science Ph.D. from an Ivy League School: Today, more than ever before, you can access cutting-edge technology and state-of-the-art math just by importing some libraries into your Python script (Java, C++, Julia, R, etc. can also do it).
  • You need to know about “Big Data”: Thanks to Transfer Learning techniques (When you have a network trained on a huge dataset, then you fine-tune it to your smaller dataset), you can, relatively quickly, set up your own Neural Network especially suited for your application. It has been proven to be very successful. The reason is that the lower -early- layers are generalizable features.
  • It would be best to have expensive computational power: Most practical, real-world projects we work on or know about required 1 GPU. You can rent 1 GPU for 45 cents per hour. If you need to train a complex model on a decent amount of data (let say 70 hours on training and testing), it only will cost you about $31.5.
  • You have to spend years studying advanced math before you can use AI: As stated before, many initiatives have made it easier to have access to cutting-edge algorithms and techniques with little or no knowledge at all on advanced math.
Interesting for you:  Why should you be skeptical when it comes to hyped-up AI?

To read more on Artificial Intelligence, click here.

#ArtificialIntelligence #AIDEMOCRATIZATION #BenefitsofAI #AI #BusinessDevelopment2021

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

%d bloggers like this: