The Top 4 Risks Of Outsourcing Gen AI To China

 

While there seems to be no end to the tariff war in China, many top CEOs are warning the Administration that there could very well be empty shelves in the major grocery stores, and other related goods stores here in the United States.  In fact, even the major shipping containers that are coming from China are now starting to slow down.

Because of this, many countries could very well be turning to China now to be the major trading partner, replacing the US entirely.  One such area in which this is happening is in the Generative AI Industry. We have already seen this with Nvidia, where severe restrictions are now being placed upon them onto the kinds of chips that they can export there.

But one area which people could very well turn to China is around actually developing the models that drive Generative AI.  After all, why pay more here in the US when you can have the same thing done there faster and cheaper (but of course, the quality of the development will still be an issue.

But there are inherent risks depending upon another country to do this.  Here are some of them:

1)     Biasness :

The technical definition of Generative AI biasness is:

“Artificial intelligence bias, or AI bias, refers to systematic discrimination embedded within AI systems that can reinforce existing biases, and amplify discrimination, prejudice, and stereotyping.”

(SOURCE :  https://www.sap.com/resources/what-is-ai-bias)

To put it in another way, this is when the out output that has been yielded by the model produces some kind of content that is deemed to biased, or even racial in some way.  Although this is a direct product of the datasets that have been fed into the model, a good Gen AI programmer could still tweak the algorithms, so that they can still produce this same kind of content, even though the data might have checked before time. 

2)     Optimization:

In the world of Generative AI, this is also known as “fine tuning”.  This is where you are trying to keep all the models in top condition so that they produce the best possible outputs.  Obviously, if you have created the model, you will know immediately how to do this.  But what if  you had outsourced the model creation to another company in China?  Obviously, they are not going to tell the secret sauce to their recipe, so fine tuning here could be a major problem, because you will not know the inner workings of the model.

3)     A Deepfake?

A Deepfake, as its name implies, is a “fake” version of a real person.  This is quite widely used during the political election seasons, where a Cyberattacker could post a fake video of a politician asking for donations to their respective campaign.  So, in this regard, how do you know that a Generative AI model that has been developed for you is not the real thing?  What if you are just getting a “Deepfake” of it.  This is an especially worrisome situation, since your customers will also be inputting data and information into the submittal forms of your web application.  This in turn will also be fed into your Gen AI Model, so that you can analyze any trends to help you determine the viability of new products and services.

4)     The Creation:

Whenever you hire an outside source to develop your Gen Models, you will also want to meet the team that will be doing, whether it is virtual or face to face.  Be very leery of hiring a company from overseas that does not introduce their team to you.  After all, it could be a Cyberattacker that is creating it and could put all kinds of covert backdoors into the code so that they can gain direct access to your IT and Network Infrastructure. 

My Thoughts on This:

The risks that I have described here can not only happen in China,  but it could even very well happen here in the US.  The key difference is that we contract in place that can be enforced in a court of law, though it may take some time. 

If you choose to outsource this to a company, say once again to China, and they violate the terms of the contract that they have signed with you, it will be very difficult best, if not impossible, to gain any kind of legal recourse.

So, while faster and cheaper might be the way to go, think twice about that.  Quality will always beat those two in the end, no matter what the need or the application is.