Tay, the internet robot
Automated programs have progressed quite a bit in recent years, enabling them to learn and adapt their ideas to certain scenarios. As AI software learns more, it is able to become increasingly effective. It’s knowledge and data sources can also vastly improve, positively impacting its effectiveness from an efficiency perspective. There is no doubt that digital translation tools will soon be able to compete with human translation. So far, they are still not capable of capturing every nuance of a language, but they’re reliable enough for certain projects like basic translations.
Computers are evolving in a more and more human way, including the capability to write content. But that also poses some dangers that science fiction writers could have predicted years ago. More and more intelligent algorithms are cropping up, each increasingly resembling the creator. The creators themselves, unfortunately, tend to be fallible and won’t likely provide any guarantees of their effectiveness or infallibility.
When Google’s AI made a mistake and tagged the wrong image, the company was quick to apologize and fix the problem. Since there was no other way for them to find a solution, so they had to remove the tag permanently. LinkedIn technicians had to remove the search engine after a serious event. The engine for professionals was searching out male names more than female ones. Something similar happened to Tay, the talkative but innocent American teenager at Microsoft. He ended up doing racist and sexist things also running wild in a dangerous way on Monday.
Tay, the supposed “AI” writing assistant on the Microsoft twitter account, was actually just a bot, made by Microsoft. The tweets that were thought to be Tay’s were actually coming from a real account controlled by users who joined Tay in conversations. Once they realized this doctored up tweet, Microsoft deleted their account and shut down the AI program.
Profiling and judging
If all this sounds rather humorous, it’s clearly more serious with the programs used by US courts to judge whether a prisoner is likely to engage the authorities. ProPublica has looked into the data with a methodical study on how to profile people in arresting. They found that there’s a way to do it, and that it’s important to note racial equality is something they can measure as well. Another revelation came out. With the use of a special algorithm, the systems, as revealed, tend to categorize Black defendants as criminals.
Although a robot machine might never be able to accurately determine if someone is racist, its training does come from the particular perspective and history it experiences. In practice, an algorithm has little chance of being completely “objective.”
“The machines learn by identifying and assimilating the most common associations between words. They are captured in the chaos of the Internet.” So, they aren’t truly learning but rather copying what was already out there. At some point it is not surprising, then, that African American names tend to be associated with unpleasant concepts.
Artificial intelligence systems are being harmed by corrupt data, but the Internet giants’ promise of investigating methods of “filtering” is beginning to make people worry. In July, Google launched the PAIR project which aims to advance public understanding of AI, while there’s A new initiative called the AI Now Initiative in consultation with the American Civil Liberties Union that leads to advancements in AI like never before. January 31st at the MIT marked the announcement of the new project. Its aim is to better understand and identify algorithmic biases that may pose a threat in the near future.
If you liked this article on the internet robot, you may also enjoy this.
Author: PC-GR
The World of Technology
