Serial No

Statement

Origin

Why incorrect

Corrected version

1,

AI algorithms do not scale up to address real world domains in robotics and language processing

Lighthill Report (1973) by James Lighthill.

The fact that we have advanced robotic inventions such as self-driving vehicles and autonomous drones in operation today and large language model solutions such as ChatGPT and Google PaLM2 in circulation,  proves that Lighthill’s pessimistic assertion 50 years ago was clearly an entirely  wrong one.

There has been much achievement and product development in the areas of robotics and natural language processing by applying AI/ML algorithms.

2.

We do not really understand how deep learning works and how machines learn

Various individuals and groups

Machine learning algorithms are based on advanced mathematics which requires a specialized knowledge background to comprehend them thoroughly. A number of individuals who do not have the background to understand the workings of these advanced systems end up stating that we cannot really understand them.

Every machine learning system/algorithm is based on advanced mathematical models which are human designed and encoded hence we do know exactly how they work.

3.

Machines will develop the capability to design other machines which will eventually take over the world and destroy humans.

Stephen Hawking, late English cosmologist and physicist

Professor Hawking who was seriously ill for many years made this statement just before he passed away hence his prediction should not be considered seriously.

Although some automatic coding capabilities are now in place we are nowhere neardesigning machines that can design other machines, let alone make autonomous decisions to kill people.

4.

Enterprise AI agents are off-the-shelf plug and play  products which can learn on their own and make decisions

Marketing teams presenting misleading hype

To be effective Enterprise AI systems require to be trained on business domain data before they can be used to predict, detect optimize and recommend. Such data requires to be engineered and pre-processed prior to being fed to an AI/ML agent. Periodic retraining is necessary for the levels of effectiveness to be retained and improved. Thus AI solutions for enterprises being domain specific, are not plug and play in nature.

Enterprise AI agents must be customized within the intended business domain before they can be truly effective in augmenting human decision making.

5.

AI will lead to massive job reduction, unemployment and strife within society.

Various pessimists some of whom may be influenced by highly fanciful video game creations depicting androids taking over various human work functions which is very far from today’s reality.

AI is today a multi-billion dollar industry which has already led to the creation of thousands of new jobs. As per The World Economic Forum’s ‘The Future of Jobs Report 2020’ AI can potentially generate 97 million new roles in the next 5 years.

Although a certain number of manually executed jobs will be replaced by newly introduced methods of automation the introduction and growth of new technology such as AI/ML requiring new skillsets has already created thousands of new job opportunities.

6.

AI-based systems produce biased results and are unreliable.

Sponsors of failed AI projects

It is essential for a machine learning model to strike a balance between bias and variance in order to produce accurate results. A machine learning model also needs to be adequately verified with test data before being  deployed for business use. Not following such best practices will lead to an unreliable model. 

Best practices of model design and test data based verification are both essential factors for an AI system to be built for reliable operation.

7.

AI works exactly like a human brain 

A general common misconception which is widespread.

While AI systems have been developed by researchers attempting to emulate the reasoning capabilities of the human brain, the AI systems that we have developed today are incapable of replicating many of the complex and  advanced abilities of the human brain such as intuition and instinct.

While AI can mimic certain aspects of human cognition, it is not capable of replicating the complex advanced  abilities of the human brain.

8.

AI will lead to the creation of systems which will autonomously decide to kill people.

Science Fiction books and films

The fanciful notion of machines who decide autonomously to kill humans is pure science fiction. Some well-known examples are provided in the next section. Unfortunately the impact of such science fiction movies on people’s minds is rather significant. Nonetheless regulatory measures are needed for AI and all advanced technology to avert the danger of extreme misuse.

Using AI for nefarious objectives can be only human driven and not  autonomously perpetrated by a machine however intelligent it might be. This is because AI-based systems are human-designed and can do only what they are trained to do. However the risk of misuse of the intelligent technology for harmful intent is indeed a grave concern hence regulatory measures for designing AI systems are essential.