Should We Worry About Generative AI? Here Are 12 Reasons Why We Should

We’re just now starting to fully understand what generative AI is capable of. Several people now assert that generative AI models, such as ChatGPT, are not just on par with humans in intelligence but are often even smarter than them. They have writings that are rich in information, concepts, and knowledge. They produce stunning artwork in a bewildering variety of genres. It’s difficult to believe the manufactured artifacts are the work of a computer since they are so diverse and seem to be one of a kind.

Some onlookers prefer to believe that these emerging AIs have successfully passed the Turing test. Some think the barrier has been blasted apart rather than crossed peacefully. Another group is undoubtedly on their way to the unemployment line since this art is so excellent.

Yet as soon as the sensation of amazement wears off, generative AI loses some of its star power. Some observers have done a sport out of asking questions in such a manner as to get stupid or false information from artificially intelligent robots. Some use the tried-and-true logic bombs from elementary school art class, such as requesting a painting of a polar bear in a blizzard or the sun at night. Others generate odd demands that highlight the limitations of context awareness, or common sense, in AI. Anybody with the urge may list the ways that generative AI falls short.

The drawbacks and flaws of generative AI are listed below. The envious writing of a writer who stands to lose their job if the computers take over may make this list seem like bitter grapes.

Plagiarism

In reality, generative AI models like ChatGPT and DALL-E are merely creating new patterns from the enormous samples in their training set. The end product is a cut-and-paste synthesis taken from numerous sources, sometimes called plagiarism when done by people.

Humans do learn by imitation, after all, but sometimes the borrowing is so evident that a teacher in a primary school would recognize it. Such artificial intelligence-generated material comprises lengthy passages of text that are given almost verbatim. Yet sometimes, there is so much blending or synthesis going on that even a panel of academics from a college could struggle to identify the source. In either case, originality is lacking. These devices, for all their brilliance, cannot create anything genuinely original.

Copyright

Although copyright laws are applicable in the marketplace, plagiarism is mostly a problem in schools. When one person steals from another’s labor, they run the possibility of being prosecuted and facing penalties of millions of dollars. What about AIs, though? Do they fall under the same laws as us?

It will take years to resolve the legal position of generative AI since copyright law is a complex topic. But remember that some people will undoubtedly use their newfound free time to file lawsuits if AIs generate work that seems excellent enough to place humans on the job market.

Work performed for no pay

Besides copyright and plagiarism, generative AI also raises other legal concerns. New ethical disputes are already being proposed by lawyers for court cases. Should, for instance, a corporation that creates drawing software be permitted to gather information about how human users draw and use that information to teach AI? Should such creative effort be paid for by humans? A key element in the success of the current generation of AIs is access to data. What happens then if the data producers want a piece of the action? Which is just? What should be deemed lawful?

Information is not knowledge

The type of intellect that takes years to acquire in humans is especially well mimicked by AIs. We have an excellent right to be pleased when a human researcher can present a little-known 17th-century artist or compose new music using an almost extinct renaissance tonal structure. We are aware that acquiring such an amount of information requires many years of study. The outcomes may be dazzlingly accurate and right when an AI does these identical tasks after just a few months of training, yet something is still wrong.

AIs may seem to mimic human creativity’s unpredictable and whimsical aspects, but they cannot do so. A well-trained computer can discover everything there is to know about a poet like Aphra Behn if it can find the relevant old receipt in a digital shoebox full of billions of data. Mayan hieroglyphics may even have been translated by technology. Yet, creativity and innovation are driven by unpredictability. Businesses like fashion are characterized by change and are hooked to it. In actuality, both traditional human intellect and artificial intelligence have their role.

Intellectual sluggishness

Speaking about intelligence, AIs are by nature rule-based and mechanistic. An AI builds a model after slogging through a training data set, and that model doesn’t truly alter. Some data scientists and engineers envision progressively retraining AI models over time so that computers may pick up new skills. But in general, the goal is to develop a sophisticated network of neurons that permanently record specific information. Constancy has been used and may be effective for certain sectors. The risk with AI is that it will always get enmeshed in the culture of its training data. What happens if humans grow so reliant on generative AI that we cannot create new data to feed the models?

Security and privacy

There must be a source of training data for AIs, but what gets locked within the neural networks sometimes needs to be clarified. What if AIs misuse the training data they have access to? To make things worse, since AIs are created to be so adaptable, locking them down is far more difficult. Access to a specific table containing personal information may be restricted using a relational database. Yet, there are several methods to query an AI. Attackers will rapidly pick up on the proper techniques for asking relevant questions to get the sensitive information they want. Consider the case when a certain asset’s latitude and longitude are restricted. A cunning attacker may request the precise time the sun rises in that area over many weeks. An obedient AI will attempt to respond. We have yet to learn how to teach an AI to secure sensitive data.

Undiscovered prejudice

Insufficient training data is the cause of many AI issues. The outcomes will be skewed or erroneous if the data set is. GIGO, “garbage in, garbage out,” was created by one of the original mainframe programmers, who recognized the fundamental issue with computers.

While the generative AI technology may be as logic-driven as Spock, the people who create and teach the machines are not. It has been shown that bias and partisanship make their way into AI models. Possibly biased data were utilized to build the model. They could have implemented overrides to stop the model from providing some controversial answers. They could have pre-programmed the responses, making them difficult to spot. Humans have developed several strategies to guarantee that AIs are great platforms for our toxic views.

Automation stupidity

Since AI models are so good at many other things, it’s simple to overlook their flaws. Many errors are hard to predict since AIs think differently than we do. For instance, many users of text-to-image tools have discovered that AIs often need to pay more attention to basic tasks like counting. Basic math is a skill humans pick up in elementary school and apply in many situations. When drawing an octopus, a 10-year-old will almost likely ensure it has eight legs. The abstract and contextual applications of arithmetic often cause the most recent iterations of AI to stumble. If model makers focused on the flaw, this may easily alter, but there will still be other flaws. Artificial intelligence is distinct from human intellect, so artificial stupidity will also be different.

Human credulity

We humans often fill the gaps in Artificial intelligence without recognizing it. We interpolate solutions or add missing details. We don’t dispute it if the AI says that Henry VIII was the monarch who murdered his wives since we aren’t familiar with that period of history. When a likable presenter swings their hands, we automatically believe that the AI is accurate. The human mind thinks anything is real and precise if said confidently.

The hardest challenge facing generative AI users is recognizing when the AI is in error. Since they can’t lie like people can, machines are much more harmful. They may provide paragraphs of absolutely factual information before straying into conjecture or blatant defamation without anybody noticing. Used car salespeople or poker players often have a tale that makes it obvious when they are faking; AIs don’t.

Endless abundance

The endlessly replicable nature of digital material has already put a strain on many economic theories based on scarcity. These models will be even more broken by generative AIs. Some authors and artists may lose jobs due to generative AI, which also changes many economic norms that govern our daily lives. When both the advertising and the content can be merged and created indefinitely, will ad-supported content function? Will the free sector of the internet become a place where generative AIs create and endlessly reproduce bots that click on website ads?

Such simple plenty has the potential to harm the whole economy. If non-fungible tokens can be duplicated endlessly, would people still buy them? Will art still be recognized if it’s so simple to create it? Will it continue to be unique? Will it matter if it’s not unusual? When everything is taken for granted, does anything still have value? Was this what Shakespeare meant when he talked of the outrageous fortune’s slings and arrows? Let’s not attempt to provide an explanation for it. Asking a creative AI for a hilarious, peculiar, and ultimately mystifyingly imprisoned response to a question about right and wrong is a good start.

Accurate data and quality assurance are required

The fact that the data or content a certain model creates or produces still has to be examined is one of generative AI’s key drawbacks. There are various occasions when a certain model has generated low-quality outputs, which include inaccurate findings, a lack of relevance, and other issues.

The best illustration is ChatGPT. It is unable to respond to questions concerning more recent occurrences. Moreover, it results in confusing and repeating responses to specific inquiries. The James Webb Space Telescope was incorrectly said to have been used in advertising by Google AdWords, producing the first images of a planet outside the Solar System.

Remembering that the caliber of the datasets or training sets determines the quality of the outputs generated by a certain generative model. Biases existing in the training data may be reflected in a particular model. If the training set is skewed, this might lead to biased findings. Both the output quality and dependability are impacted by these findings.

Challenges in Complexity and Technology

Another drawback is that the technology behind Generative AI might be challenging to comprehend. Many individuals need to understand how it operates or how to utilize or use it. This works against its benefits and uses. Since it is a complicated and new technology, a small firm can be unwilling to employ it.

Premium services provide better security and adaptability. There are restrictions with free services like ChatGPT and Dall-E. For instance, ChatGPT often experiences downtimes during periods of high use. Dall-E is likewise cost-free. However, each user is only permitted to create up to 50 free photographs in the first month and an additional 15 images afterward.