Assignment 2 - Case Analysis thtutor 1041

Google’s Bard mistake (AI misinformation) (no hyperlink, this was widely covered in the media) 1 | P a g e

Table of Contents Introduction ................................................................................................................................ 3 Discussion .................................................................................................................................. 3 Crisis typology ....................................................................................................................... 3 Media visibility ...................................................................................................................... 5 Stakeholder impact ................................................................................................................. 7 Crisis response ....................................................................................................................... 8 Recommendation ........................................................................................................................ 8 Conclusion .................................................................................................................................. 9 References ................................................................................................................................ 11 2 | P a g e

intent, building on Google understands of high-quality information. However, like all LLMs, Bard has the potential to produce responses that are erroneous or deceptive even when researcher presents himself as confident and persuasive. Since LLM's fundamental working principle is to anticipate the next word or sentence, it currently struggles to discern between true and false information ( Byrne, 2023). For instance, if Company ask an LLM to solve a textual maths issue, instead of using sophisticated reasoning and calculation, the LLM will forecast the solution based on other learnt insights. To that end, we've seen Byrd give explanations that are either made up or false. Bias Training data includes viewpoints and ideas from a range of sources, including the general population. Google keep looking for new ways to apply this information so that LLM responses can take a greater diversity of viewpoints into account while avoiding unfavourable reactions. When attempting to anticipate suitable responses, the model may include any gaps, biases, or prejudices included in the training data in its output. These problems can show themselves in a variety of ways, such as comments that solely reflect one culture or demographic, responses that make use of unfavourable stereotypes, or responses that demonstrate gender, religious, or racial discrimination ( Dankova, 2023). The information is incomplete for several subjects. In other words, there isn't enough trustworthy knowledge available for LLM to research a subject thoroughly and develop accurate conclusions. In these situations, there is an increase in the production of false or erroneous information. Everyone's safety is a priority when building a safe environment at Bard, and this is an ongoing goal. Through constant fine-tuning Google enhance Bard's training data and system. In order to conduct research and create domain roadmaps with in-depth topic experience outside of Google, google also collaborate with a variety of communities and domain experts. False positives / negatives We have put in place a number of technical protections to stop Bard from reacting to impromptu requests or showing objectionable or hazardous content. These guardrails are meant to stop undesirable reactions, but Bird has the ability to misinterpret them, leading to "false positive" and "false negative" outcomes. If a "false positive result" happens, Bard might not react appropriately or might perceive the cues incorrectly. The safeguards in place might not prevent a "false denial," and Bard might still elicit an unwarranted response. In order to comprehend and categorise safe entry and departure places better,googlekeep improving these models ( Dwivedi et al., 2023). As civilizations, events, and languages change quickly, this scenario will persist. 4 | P a g e

Media visibility Google Ads can assess whether an advertisement is displayed to a potential customer using the Active View technology found on YouTube and some websites and applications in the Display Network. For video and display campaigns, use the Active View measure to track how frequently ads appear on websites, mobile devices, and apps. The possibility that a user will find a specific website when looking up related keywords is measured by search visibility. This is determined by the volume of searches, the number of keywords a website ranks for, and the position of the domain in the search results ( Fraiwan and Khasawneh, 2023). In order to evaluate website prospective traffic from organic search results to that of rivals, it is crucial to analyse this measure. In order to evaluate website's prospective traffic from organic search results to that of rivals, it is crucial to analyses this measure. Low or no website visibility indicates that consumers cannot reach website through organic search and that domain performs poorly for the most pertinent keywords. On the other hand, a website is said to have great exposure in search engines if it gets high SERP ranks for a lot of (popular) keywords. The maximum visibility is given to websites that are among the top three in the SERPs. A domain's visibility in search engines is very low if it does not show up on the first page (top 10) of results ( Garon, 2023). 75% of Hubspot users, according to study, never scroll past the first page of Google search results. As a result, the first page of website receives the most visibility and consequently the most visitors. In the study, More Sistrix discovered that researcher received 28.5% of his clicks from Google's first organic search result, 15.7% from the second, and 11.0% from the third. How to use Bard As was previously noted, Bard is capable of producing unique answers even to queries and prompts that are the same or quite similar. In initial testing, Google found that it was helpful for users to be able to view some of these many responses, particularly for creative topics like poetry or short tales or when there wasn't a single right answer. Researcher can confirm that if the user chooses "Show other draughts," they can view many versions of the bard's response and pick the one they want ( Morris, 2023). Bard is made to produce unique output based on the underlying prediction method, just like other independent LLM-based interfaces. In rare circumstances, the response may include references to previously published material. If Bard takes a straight quote from a website, researcher will cite it so that readers can find the page 5 | P a g e

features, the problem was discovered. Stakeholder impact According to reports, Alphabet's shares dropped 8% (8.59% per share) to $99.50, making it one of the most frequently traded businesses on American stock exchanges. The absence of information during Google's AI search event, which failed to demonstrate how Microsoft's ChatGPT challenge will be solved, was blamed by analysts for the decline in market value. The incident emphasises the significance of meticulous quality assurance and testing procedures in creating and releasing AI-powered products. Google acknowledged its error and made the announcement that it will launch a programme called Trusted Testers that combines external feedback with internal testing to make sure Bard's responses adhere to its strict standards for quality, security, and real-world information ( Taecharungroj, 2023). The decrease in Alphabet's market value serves as a warning that AI technology is not absolute and that it can have important ramifications for businesses and their stakeholders. Organisations must prioritise creating effective testing and quality assurance processes as AI continues to change how humans interact with information in order to prevent expensive errors and reputational harm. In conclusion, his loss of more than $100 billion in market value for Google emphasises the necessity for accuracy and care while developing and releasing AI-powered goods. To ensure that his AI products are of the best calibre and are trusted by users as AI develops, the organisation must place a strong priority on quality and security. To determine how AI will affect the future of business, the online research platform Real Research has launched a study of Google, which lost $100 billion in shares due to AI gaffes. A vote on his $100 billion Google loss caused by AI chatbot defects will be conducted on February 17, 2023, on the Real Research app ( Wandhöfer and Nakib, 2023). If researcher participates, researcher will earn 60 multinational companies as a prize. According to Bloomberg, 18 current and former Google employees have disseminated inaccurate information, showing that Byrd "deprioritizes his ethical obligations" and "does not lose out to the competition." researcher claimed to have viewed an internal document that said that. According to screenshots of internal conversations obtained by Bloomberg, Google employees referred to Byrd as a "pathological liar." Another employee referred to it as "terrifying." That's not all, though. Mr. Bird supplied advice that caused the crash when a worker approached him for guidance on how to land the aircraft. According to a different official's letter, Mr. Byrd's diving-related responses would probably cause "injury or 7 | P a g e

death( Grünebaum et al., 2023)." According to the Bloomberg story, personnel who are in charge of new product safety and ethics have been told not to "interfere with or attempt to destroy generative AI tools under development." Crisis response Ongoing research and development Bard advances Google's most recent LLM efforts, which include the launch of the Natural Conversational Model in 2015. With the use of this framework, it was shown how the model can predict the subsequent sentence in a discussion based on the preceding one, leading to a more natural conversational experience. A crucial task for comprehending natural language and artificial intelligence is conversation modelling. There are earlier methods, but they frequently have limitations (such being only applicable to booking airline tickets) and demand hand-crafted regulations ( Khorashadizadeh et al., 2023). In this piece, use the recently proposed Sequence-to-Sequence His architecture to give a simple method for doing this operation. Approach changes by foretelling the subsequent statement in a discussion using the prior sentences. model's strength is that it can be taught from beginning to end with much fewer manually created rules. It turns out that with a sizable conversational training dataset, this straightforward model can produce straightforward talks. Application of AI Principles Accountability and safety are the two pillars on which everything google do at Bard is built.Company help Bard flourish by, among other people, delivering a huge social benefit. review of the early and promising Bard applications in need of a resilient web content ecosystem is now complete. Google are dedicated to responsible innovation in this space, which includes collaborating with content producers to determine how this new technology may enhance their work and benefit the broader online ecosystem. Accountability and safety are the two pillars on which everything Google do at Bard is built( Rubin, 2022). Google are dedicated to responsible innovation in this space, which includes collaborating with content producers to determine how this new technology may enhance their work and benefit the broader online ecosystem. We're still working on this as google develop Bard, but AI principles also emphasize the need to prevent harm. Google routinely stress test models with internal members of "red team" (product experts and social scientists), looking for flaws, assessing fairness and gender concerns, and assessing business potential. Recommendation  Google requests that all interactions with Mr. Bard be "polite, friendly, and 8 | P a g e

References Akter, S., Hossain, M.A., Sajib, S., Sultana, S., Rahman, M., Vrontis, D. and McCarthy, G., 2023. A framework for AI-powered service innovation capability: Review and agenda for future research. Technovation , 125 , p.102768. Awad, A.G., 2023. Can Artificial Intelligence and Big Data Analytics Save the Future of Psychiatry?: The Search for a New Psychiatry and Other Challenges . iUniverse. Byrne, M.D., 2023. Generative Artificial Intelligence and ChatGPT. Journal of PeriAnesthesia Nursing . Dankova, B., 2023.Company Had Better Check the Facts: Reader Agency in the Identification of Machine-Generated Medical Fake News. Reinvention: an International Journal of Undergraduate Research , 16 (1). Dwivedi, Y.K., Kshetri, N., Hughes, L., Slade, E.L., Jeyaraj, A., Kar, A.K., Baabdullah, A.M., Koohang, A., Raghavan, V., Ahuja, M. and Albanna, H., 2023. “So what if ChatGPT wrote it?” Multidisciplinary perspectives on opportunities, challenges and implications of generative conversational AI for research, practice and policy. International Journal of Information Management , 71 , p.102642. Fraiwan, M. and Khasawneh, N., 2023. A Review of ChatGPT Applications in Education, Marketing, Software Engineering, and Healthcare: Benefits, Drawbacks, and Research Directions. arXiv preprint arXiv:2305.00237 . Garon, J., 2023. A Practical Introduction to Generative AI, Synthetic Media, and the Messages Found in the Latest Medium. Synthetic Media, and the Messages Found in the Latest Medium (March 14, 2023) . Morris, M.R., 2023. Scientists' Perspectives on the Potential for Generative AI in their Fields. arXiv preprint arXiv:2304.01420 . Porsdam Mann, S., Earp, B.D., Nyholm, S., Danaher, J., Møller, N., Bowman-Smart, H., Hatherley, J., Koplin, J., Plozza, M., Rodger, D. and Treit, P.V., 2023. Generative AI entails a credit–blame asymmetry. Nature Machine Intelligence , pp.1-4. Rudolph, J., Tan, S. and Tan, S., 2023. War of the chatbots: Bard, Bing Chat, ChatGPT, Ernie and beyond. The new AI gold rush and its impact on higher education. Journal of Applied Learning and Teaching , 6 (1). RudolphA, J., 2023. Journal of Applied Learning & Teaching. Journal of Applied Learning & Teaching , 6 (1). Steinhoff, J., 2023. AI ethics as subordinated innovation network. AI & SOCIETY , pp.1-13. Taecharungroj, V., 2023. “What Can ChatGPT Do?” Analyzing Early Reactions to the 10 | P a g e

Innovative AI Chatbot on Twitter. Big Data and Cognitive Computing , 7 (1), p.35. Wandhöfer, R. and Nakib, H.D., 2023. The Universe of Technology. In Redecentralisation: Building the Digital Financial Ecosystem (pp. 39-74). Cham: Springer International Publishing. Grünebaum, A., Chervenak, J., Pollet, S.L., Katz, A. and Chervenak, F.A., 2023. The exciting potential for ChatGPT in obstetrics and gynecology. American Journal of Obstetrics and Gynecology . Khorashadizadeh, H., Mihindukulasooriya, N., Tiwari, S., Groppe, J. and Groppe, S., 2023. Exploring In-Context Learning Capabilities of Foundation Models for Generating Knowledge Graphs from Text. arXiv preprint arXiv:2305.08804 . Rubin, V.L., 2022. Investigation in Law Enforcement, Journalism, and Sciences. In Misinformation and Disinformation: Detecting Fakes with the Eye and AI (pp. 123-156). Cham: Springer International Publishing. Jougleux, P., 2022. Hate Speech, Fake News, and the Moderation Problem. In Facebook and the (EU) Law: How the Social Network Reshaped the Legal Framework (pp. 183-212). Cham: Springer International Publishing. Sarel, R., 2023. Restraining ChatGPT. Available at SSRN 4354486 . Shur-Ofry, M., 2023. Multiplicity as an AI Governance Principle. Available at SSRN 4444354 . 11 | P a g e