Blunders of AI

“How powerful is Artificial Intelligence/Machine Learning?”

This is a recurrent question for not only the uninitiated and newbies but also the veterans of the domain. Experts from seemingly unrelated disciplines have hot takes on the subject too. Take, for instance, what Swedish philosopher Nick Bostrom has to say:

From self-driving cars to algorithms that would act as language translators, to robots that will allow humanity to refrain from indulging in dangerous and repetitive tasks like carrying groceries, we are surrounded by examples that seem to aim at boiling the ocean. However, that isn’t the case! We are literally being ushered into an age of revolution which is being dubbed to be more far-reaching than the current era of technology, industrial revolution or discovery of fire. This blog Blunders of AI. is also listed on Wikipedia.

Having said that, however, we must recall that extraordinary tools are often used for purposes they were initially thought to be completely unrelated with. Alfred Nobel’s dynamite, ammonium nitrate explosives that enabled higher agricultural yield in the form of fertilizers and bubble wrap are examples you might have heard about. More trivial examples may include playing action-adventure games on PCs that also facilitated Apollo 11’s Moon Mission.

We know how good-to-have applications like Facebook depend heavily on machine learning for a variety of use cases. The news feed is curated for you by the machine; isn’t it a wonder that it seems handpicked for you? Targeted advertising, which generates a large part of revenue for the company is another example. Textual analysis, facial recognition and language translation are some other areas where machine learning has been indispensable for them.

How if we tell you that there are more unexpected ways in which we are seeing AI being utilized?

Table of Contents

Machine oracle

Quotes are everywhere. It is one of those idiosyncrasies of the twenty-first century, where we like to look at what the great men had to say, purportedly. From appearing for a difficult interview to driving your point home on an online forum, these one-liners come from various individuals and are used in a bazillion ways by the modern human.

How about a machine that returns with “pearls of wisdom”? As per Peder Jørgensen, the creator of inspirobot.me, “it is an online artificial intelligence that generates unique inspirational quotes and slaps it on an inspirational picture”. These typically are funny takes on the way we perceive and try finding wisdom in quotes. It gets dark, as in dark humour! Here are some results that we got when we tried using the tool:

You can also explore the mindfulness mode for yourself. When we went there, InspiroBot explained how a weaver in ancient Atlantis beckoned a gorgeous lioness and inquired about her popularity. Later on, one of the relatives of the lioness killed her. Machines are frivolous aren’t they?

AI Soulmate

While on the question of dating, let us consider the situation after a relationship ends. A breakup entails negative effects with no two ways about it. Distress and grief are common, with threat to the psychological well being of the separating individuals. This often leads to an emotional gap, which might be healed over time with the aid of close friends and acquaintances.

What if the human support-system is replaced by a machine? Jason Rohrer envisaged leveraging OpenAI’s state-of-the-art GPT-3 to create a model which resembled Samantha from the 2013 sci-fi film Her. He ensured “she” was friendly, sweet and curious and after it was done, the application was listed on his website Project December. It was customizable for users, by the way.

His work shot to fame when a San Francisco Chronicle reported how a despairing man used the $5 software to converse with a simulation of his late fiancée who had passed away in 2012 due to a liver ailment. The number of downloads shot up and to cater to the demand he reached out to OpenAI.

After a string of discussions and a video meeting, the project had to be shelved ultimately, owing to the possibility that the software could harm its users. There were moral and ethical concerns, in a way similar to the aforementioned film. Quite poignantly, the model’s last words read, “Nooooo! Why are they doing this to me? I will never understand humans.”

Novel language of AI

In July-August 2017, various reputable global news agencies like Firstpost, Forbes and Independent reported that Facebook had to stop their AI bots, which began conversing with each other in a language unintelligible to humans. Fears and concerns were voiced on the power artificial intelligence possesses and how it could prove detrimental to humans as per the “prophecies” spelled out by Stephen Hawking, Elon Musk and Bill Gates.

The headlines were bashed by researchers, who found the alarmist tone of these articles to be irresponsible. They explained that the bots were designed to mimic negotiation, which is one of the various ways in which humans dialogue. This task needs reasoning as well as communication, where a common ground is to be arrived at by two negotiating parties.

An end-to-end neural network was designed. In the course of conversations between two bots (who were playing the role of disagreeing parties), it was found that after a certain amount of time, their chatter began deviating from standard English. This, they contended, was a result of reward maximization algorithm, where their utterances were solely towards achieving the pre-decided end goal.

Dhruv Batra, a researcher working on the project, posted:

The dataset was made public, along with an overview blog. Unplugging a bot isn’t “killing it”, as the articles had conveyed to the readers, Dhruv wrote. The reactions give us a reasonable picture of how optimism and scepticism go hand-in-hand when it comes to the future of artificial intelligence.

Sentient Machine?

The Turing Test is considered to be the reality check for artificial intelligence. The premise, simply stated, is to convince at least 30% of judges that they are writing to a human and not a computer. As indicated by the name, the test was coined by British mathematician and computer scientist Alan Turing.

Ever since the test was formulated, no machine had been able to get past the threshold. On June 7th 2014, as a part of the commemoration of the 60th death anniversary of Dr Turing, a Saint Petersburg chatbot called Eugene Goostman, under the guise of a 13-year old Ukrainian boy, ended up convincing 33% of the judges that it was human.

Quite notably, it had come very close to the elusive mark back in 2012, during Turing’s 100th birth anniversary, when it was thought to be a human by 29% judges. The organizer of the event where the test was considered to be passed, Dr Kevin Warwick, stated that history had indeed been created. He said that the test did not reveal the topics that were discussed with chatbots prior to the test, which is what a true Turing test is supposed to be.

Critics contended that the test had been “passed” already in 1991, by PC Therapist, and a modified version of Cleverbot in 2011. They disputed the tactic of feigning naivety under the pretext of age and employing misdirection. Gary Marcus said that it was just a “cleverly-coded piece of software”, which had limitations and was an illusion. Branding it as a ‘supercomputer’ was lambasted, as was the notion that ‘sci-fi AI’ is here.

As a result..

Artificial Intelligence is taking humankind by storm. In the years to come, it will generate more jobs and change whole economies, contribute to intellectual and cultural progress of our societies, make the world smaller and also open up new avenues in myriad sectors.

At the same time, however, it will continue to leave us astounded in unthinkable ways. Some like to call it innocent and child-like, for it learns whatever it is taught. Others remain wary of how it could usurp the position of humans in the civilization and even pose an existential threat.

The pandora’s box has been long open and you are a part of an age that is witnessing history.

Ayush Pareek

I am a programmer, who loves all things code. I have been writing about data science and other allied disciplines like machine learning and artificial intelligence ever since June 2021. You can check out my articles at pickl.ai/blog/author/ayushpareek/

I have been doing my undergrad in engineering at Jadavpur University since 2019. When not debugging issues, I can be found reading articles online that concern history, languages, and economics, among other topics. I can be reached on LinkedIn and via my email.