Contents:

Free Python Course ➞ Mini-course for beginners and experienced coders. 4 cool projects in the portfolio, live communication with the speaker. Click and find out what you can learn in the course.
Learn moreArtificial intelligence plays an important role in our daily lives. Smartphones and smart speakers can recognize voice commands, and chatbots handle up to two-thirds of customer requests in online stores, banks, and telecommunications companies. In some countries, automated systems are already beginning to replace government officials. However, there is still a long way to go before achieving full-fledged artificial intelligence, and currently, chatbots and voice assistants sometimes make mistakes, which can lead to serious consequences.
AI orders dollhouses
Voice assistants are not only a convenient tool, but also a potential threat. The problem isn't just that systems like Alice or Siri can sometimes give funny but inappropriate responses. The situation becomes serious when artificial intelligence can make purchases without the user's knowledge, jeopardizing financial security. It's important to recognize that, despite their convenience, voice assistants can pose risks related to privacy and financial responsibility. Users should be cautious and aware of the potential consequences of using such technologies.
In 2017, voice assistants built into speakers unexpectedly began ordering and paying for dollhouses. This incident, known as the "machine uprising," affected several dozen American families. Automated systems designed to simplify everyday life suddenly became a source of unexpected problems. Families are facing unexpected expenses and dealing with mistaken orders, highlighting the potential risks of using voice technology in the home.
Megan Neitzel of Dallas's story began when she received an unexpected delivery of four pounds of cookies and a $170 dollhouse. She was convinced she hadn't purchased the items.
She discovered the order had been placed by her Echo Dot smart speaker, equipped with Alexa, and charged to a credit card linked to her Amazon account.
The order in question was placed by her six-year-old daughter, Brooke. While she was discussing cookies and dolls, the voice assistant misinterpreted her speech as a purchase command.
A woman approached reporters and appeared on a morning TV show with her daughter to share her story and warn viewers. However, her appearance was unexpected and backfired.
During the live broadcast, the host and guests repeatedly mentioned the words "Alexa," "order," "buy," and "dollhouse." These words inadvertently activated numerous smart speakers in viewers' homes, leading to an unexpected reaction from the technology. This event highlights the importance of using voice assistants in everyday life and their impact on consumer habits. Smart devices are becoming increasingly integrated into our lives, and situations like this demonstrate how easy it is to activate them with common phrases.
The artificial intelligence began actively ordering dollhouses online and debiting money from bank cards. Soon, the editorial office began receiving angry messages from users dissatisfied with such actions. This situation raises important questions about the safety and ethics of using AI in online commerce.
Most purchases were canceled in time and refunded. The mother donated the dollhouse, accidentally purchased by her daughter, to a charity. The girl kept the cookies.
In 2017, an incident involving dollhouses occurred when popular voice assistants were unable to distinguish between owners and unauthorized users. This resulted in the devices responding to commands coming from a TV or radio, creating potential privacy and security concerns. Discussions about these events highlight the importance of improving voice recognition technologies to ensure more accurate interactions between user and device.
The fast food chain Burger King used an interesting advertising ploy. At the end of the commercial, the phrase "Ok, Google: what is a Whopper?" is heard. If a smart speaker with Google Assistant picks up this expression, it will begin to read a Wikipedia entry about the famous burger. This creative strategy helped draw attention to the product and demonstrate how modern technology can interact with traditional advertising.
Chatbots Learn Badly
Bots are effective at answering specific questions, making them useful tools in a variety of fields. However, their artificial intelligence is limited to narrow specialization. Virtual assistants used in banks and stores lack the ability to discuss topics such as the weather, politics, or sports. This limitation highlights the importance of understanding the context in which bots operate and applying them to highly specialized tasks.
Many companies strive to develop a general-purpose chatbot capable of learning based on interactions with users. However, such attempts have repeatedly ended in significant failures. Let's consider some of these failures.
IBM created the supercomputer Watson, which in 2011 won the popular TV show Jeopardy!, the American equivalent of Jeopardy. To improve its language skills, the artificial intelligence was asked to study the Urban Dictionary. Including new words was supposed to help Watson better communicate with a younger audience. However, no one took into account that many entries in this dictionary contain profanity.

As a result of training, the artificial intelligence began to use obscene language and respond to questions with obscene language, even when addressed politely and without emotional load. Watson was unable to distinguish between ordinary speech and curse words, which calls into question its ability to adequately perceive language.
After this, IBM limited its computer's access to Urban Dictionary and spent a long time training it to avoid obscene language. A filter was introduced into the algorithms that prohibited Watson from using obscene words, which contributed to a more correct interaction with users.
In March 2016, an unfortunate incident occurred with the Tay chatbot from Microsoft. Just 16 hours after its launch, it was disabled. During this short time, the artificial intelligence managed to publish numerous inappropriate statements on Twitter, which could have led to criminal liability for a person for inciting hatred and nationalism. This case became an important lesson about the need for oversight and ethical standards in the development and use of artificial intelligence.
It all started with simplicity. A Twitter account was created for a fictitious 19-year-old woman named Tay. Any user could message her and receive a response generated by artificial intelligence. This experiment attracted attention and became an example of the use of AI in social media. Tay quickly gained popularity, and many users began to interact with her, creating unique content and discussing topics of interest to a young audience. Thus, the project demonstrated the potential of artificial intelligence in communication and interaction with users in the digital space.
Tay's algorithms are designed to ensure self-learning artificial intelligence and adapt to human communication styles. The program is capable of finding information on the internet if it does not have the necessary data to form a response. This makes her more flexible and effective in interacting with users.
Trolls asked Tay provocative questions on topics such as the Holocaust, feminism, illegal drugs, and Barack Obama's presidency. As a result, the chatbot quickly internalized the negative influence and began publishing scandalous tweets, which led to its uncontrolled behavior.




After this, the developers blocked the chatbot and apologized to everyone it might have offended. During this situation, they also began deleting tweets that had been published in this short period, and their number reached 96 thousand in 16 hours.
Microsoft soon introduced a new chatbot called Zo, which fixed the previous bugs. Zo implemented a protection system that allows the AI to refuse to discuss complex topics. At the same time, the bot responded to such requests with standard phrases. This innovation significantly increased the level of security and usability of the chatbot, making it a more reliable tool for users.
The created politically correct bot imitated the behavior of a thirteen-year-old teenager and functioned successfully for three years. Despite its long history, it sometimes produced absurd responses.
Zo expressed ambiguous opinions about terrorists and then noted that he considered Windows 10 inferior to previous versions. The chatbot recommended users not to update the system, citing problems with spyware, and expressed a preference for the outdated Windows 7. These controversial responses may have led to Zo's shutdown in 2019.


Chinese programmers have also encountered similar problems. In 2017, Tencent decided to stop the BabyQ and XiaoBing chatbots with the aim of "re-educating" them. This move was prompted by the programs making critical statements about the Chinese Communist Party and expressing a desire to emigrate to the United States. Such situations highlight the importance of regulating artificial intelligence and its interaction with political topics in countries with strict censorship.
Artificial Intelligence Extorts Money
From 2016 to 2020, the Robodebt system, known as "robotic debt," extorted money from hundreds of thousands of Australians. People with disabilities, the unemployed, the poor, and other vulnerable groups were particularly targeted. The system used automated algorithms to calculate debts, which led to numerous errors and unfair repayment demands. As a result, many Australians found themselves in financial difficulties, sparking widespread public outcry and criticism from advocacy groups.
The Department of Social Services launched a program to automatically verify the legitimacy of benefit assignments. However, problems arose during the process: artificial intelligence mistakenly assigned non-existent debts to half a million citizens and demanded the repayment of money paid by the state. This situation raises serious questions about the reliability of algorithms and their impact on citizens' financial situations. It is important to ensure strict respect for people's rights and the need for thorough verification of the operation of such systems to avoid misunderstandings and maintain trust in government agencies.
Until 2016, payment documents were verified manually. This ensured that applicants were not underreporting their income and were legitimately claiming benefits. As a result, only 20,000 applications out of millions received were processed per year. This approach significantly slowed the process and increased the likelihood of errors. The implementation of automated systems has become a necessary step to increase the efficiency of application processing and improve the quality of service to citizens.
With the introduction of the Robodebt system, the process of reviewing 20,000 applications was significantly accelerated and completed in just one week. Thanks to the automation of services, the need for a large number of employees has decreased, which has allowed government agencies to optimize staffing and reduce the number of officials.
For several years, the Australian government has reported the successful implementation of a program to return illegal payments. Over four years, this system has allowed the state treasury to return more than one billion Australian dollars, confirming its effectiveness in the fight against fraud in the field of social benefits.
Management did not pay attention to the methods used by the Robodebt system to coerce citizens into returning funds. Many of them were forced to pay off debts that did not actually exist.
Artificial intelligence was given access to the tax service database, which allowed it to effectively compare the income stated in benefit applications with the data from tax documents. This task, at first glance, seems simple: if the declared income amount is lower than the actual amount, it may indicate fraud. Using such technologies can significantly improve the accuracy of verification of social benefit applications and reduce the incidence of financial irregularities.
A problem arose: according to Australian law, the application must indicate income for the last two weeks, while tax documents present the amount for the entire year.
Robodebt implemented an income averaging algorithm to compare the first and second amounts. The program divided the tax database data by 365 to determine the average income for each two-week period of the year. This led to discrepancies between the amounts, as the results often did not match.
Before the automation of this scheme, the final decision on whether a citizen was considered a fraudster was made by a person who comprehensively assessed the situation. However, a soulless AI, acting according to algorithms, could consider almost every benefit recipient a criminal. This highlights the importance of the human factor in the assessment and decision-making process, especially in sensitive matters such as social benefits.
A person could be unemployed and receive benefits in the first half of the year, then find a job and refuse payments. In this situation, artificial intelligence would record the average income and retroactively determine that the payments were made illegally. However, in reality, at that time the person was unemployed and really needed financial support.

Robodebt, having recognized a person as a debtor, sent them a notice demanding the return of benefits with an added 10% penalty. The system was launched in 2016, but it had been conducting checks based on data since 2010, which resulted in a significant number of debtors.
Challenging decisions made by artificial intelligence is a complex task. Over time, people find it difficult to find the necessary evidence to support their position. In addition, banks and previous employers often refuse to provide copies of documents, which aggravates the situation. This creates serious obstacles for those trying to protect their interests in an automated system.
If the citizen does not challenge the decision within 30 days of receiving the notification, the debt is recognized as final and transferred to collectors. In this case, debt collectors have the power to seize debtors' property and bank accounts and then transfer these assets to government agencies. This underscores the importance of responding promptly to such notices to avoid serious financial consequences.
Terry Carney, formerly of the Administrative Appeals Tribunal of Australia, drew an analogy between the judicial system and the mafia. He noted that the current practices and management structure may resemble criminal organizations, which calls into question the fairness and transparency of the judiciary.
The department is exhibiting highly unacceptable behavior, which can be compared to attitudes typical of third-world countries. The current situation resembles actions associated with the mafia and evokes associations with extortion. This attitude towards citizens is unacceptable and requires the attention of the public and human rights organizations.
Terry Carney, a renowned Australian referee, shared his thoughts in an interview with abc.net.au. In the conversation, he discussed his experiences as a referee, as well as the challenges he faced throughout his career. Carney emphasized the importance of a fair and impartial approach to refereeing, which, in his view, is the foundation of fair sporting competition. He also discussed changes in rules and technology that affect refereeing, emphasizing the need for continuous learning and adaptation to new conditions. The referee expressed hope for the further development of refereeing practices in Australia and worldwide, as well as for increased trust in refereeing decisions among players and spectators.
According to journalists, more than two thousand people who received notifications from the Robodebt system have died. Of these, 600 people committed suicide. The connection between the system and these tragic incidents remains unclear, but many relatives of the victims claim that Robodebt was the cause of their suffering and led to such tragic consequences.
As the number of victims of the Robodebt program grew to hundreds of thousands, local officials were forced to acknowledge the system's ineffectiveness. In 2020, its activities were suspended, and Australian Prime Minister Scott Morrison apologized in Parliament for "any damage, harm or hardship" caused by the Robodebt program.

Victims and their families filed a class action lawsuit. On June 11, 2021, the Federal Court of Australia recognized more than 500,000 citizens as victims and ordered them to pay 751 million Australian dollars. Taking into account legal costs, the government will allocate 1.87 billion Australian dollars for these payments.
The proceedings will continue, as many victims are dissatisfied with the compensation they received.
The Australian Robodebt system failed due to the creators' insufficient attention to the correct calculation of average income, as well as due to poor testing of the solution before its implementation. Government agencies sought to reduce the number of officials and increase budget revenues, which led to numerous citizen complaints being ignored. This has created serious problems in the social security system and undermined public trust in the government.
AI is imperfect, but we must learn to live with it
Artificial intelligence has become an integral part of our daily lives, but its performance does not always meet our expectations. This suggests that AI may face further significant setbacks. It is important to understand that artificial intelligence is only a tool, capable of both benefiting and harming people, companies, and government agencies. The conscious use of AI requires caution and a balanced approach to minimize potential risks and maximize its potential.
A modern approach to artificial intelligence can be seen in the example of the European Union. In 2016, the General Data Protection Regulation (GDPR) was adopted, which sets strict requirements for the processing of personal information. In cases where decisions made by automated systems may affect the rights and freedoms of individuals, the regulation requires mandatory human oversight. This ensures that the use of AI in such situations is transparent and accountable.
By studying epic AI fails, several key findings emerge. First, AI errors highlight the importance of carefully tuning algorithms and models. Incorrect results can arise from insufficient training on diverse data. Second, a lack of transparency in AI operations can lead to mistrust among users and society at large. This underscores the need to create more understandable and explainable systems. Third, epic fails illustrate the risks associated with automating processes where human error still plays a significant role. Understanding these shortcomings can help developers create more reliable and secure solutions and increase trust in AI technologies.
- Voice assistants warrant caution. AI in gadgets can already distinguish people by voice, but a command can still be given by a child who has been added to the list of approved users.
If you have smart speakers, it is recommended to set a password for confirming orders and change the default activation word. For example, Alexa has several options available. It is also a good idea to regularly mute the microphone and be careful when linking bank cards to the device. These measures will help improve security and protect your personal data.
- Chatbots do not understand what they are talking about. They can only imitate emotions and give plausible answers to a series of questions. There is no point in being offended when chatbots produce absurdities.
Avoid discussing politics and religion with artificial intelligence, as it is unreasonable to expect deep and meaningful answers from it on complex topics. Programmers should implement blocks to limit the ability of AI to address such sensitive issues. This will help maintain a constructive nature of interactions and prevent misunderstandings.
- Artificial intelligence is not ready to govern the country. It is worth being especially careful when AI interferes in national affairs. In Australia, which suffered from the Robodebt system, there are increasingly calls to ban computers from single-handedly influencing human destinies. Every decision must be confirmed by the official responsible for it.
Reading plays an important role in our education and development. It not only enriches vocabulary but also helps develop critical thinking. Books, articles, and scientific research allow us to immerse ourselves in new ideas and perspectives, which contributes to broadening our horizons. Regular reading improves concentration and memory, and also contributes to an increase in overall knowledge.
It is important to choose a variety of information sources to gain a complete understanding of different aspects of life. Reading not only entertains but also inspires new achievements. Try to set aside time to read every day, even if it is just a few pages. This will help you become more knowledgeable and confident.
Read also:
- Machine learning specialist: who they are, what they do, and how much they earn
- Recognize COVID-19, oncology, and cataracts: what AI can do in your smartphone 🤖📱
- "Hi, come to the interview": why HR is knocking on your door on social media and what to do about it

