Poe in Cyberspace: Looking for Poe

If you search for Poe on the Internet, you may be surprised. The first hit on Google, for example, will not be Edgar Allan Poe, as you might expect, but an expansion of the letters in the word "poe" as an anagram, first as the game Path of Exile, and later as the electrical term power over ethernet. Dig deep enough and poe will stand for portfolio of evidence and even become a unit of literary fear. In the classic movie Dr. Strangelove (1964) "poe" was the recall code for nuclear weapons. Where then is the Poe site you were looking for? Okay, let's go directly to the website called Poe. com. Another surprise: Poe. com is now an AI chat site. How did this happen? It turns out that in 1996 the Professional Office Equipment company created the Poe. com site as an anagram of its three initial letters POE. The only Poe ever to occupy the site was the novelist and journalist Richard Poe (no relation). In 2002 the Internet investor Greg Ricks bought the site for 3, 200, lost it to Richard Poe in a cybersquatting suit, bought it back in 2009 for 18, 500, and finally in 2022 sold it to Quora for 230, 000. Adam D'Angelo had founded Quora in 2009 as an interactive question-and-answer site and served as CEO. Once chief technology officer of Facebook, he joined the board of directors of OpenAI in 2018. In his dual leadership role on both Quora and OpenAI, he was in an unusual position to comment on artificial intelligence (AI): "It will probably be the most important event in the history of the world and it will happen in our lifetimes" (Twitter, August 20, 2023). Merging the qualities of Quora and OpenAI, Poe. com became a new kind of website. Poe. com claimed that its name was an anagram of Platform for Open Exploration. Although the site never credited Edgar Allan Poe, perhaps it benefitted from his qualities of independence of inquiry, deep interest in machine intelligence, and recurring passion for mathematical logic. Thus, in his article on "Maelzel's Chess-Player, " Poe was already aware of Babbage's use of machine intelligence in his pioneering computer experiments. By closely observing Maelzel's device, which had attracted great acclaim all over the world as an automaton, Poe concluded that its claims were a fraud since it had to conceal a human operator within. For Poe the giveaway was that it sometimes lost a game: "Were the machine a pure machine, this would not be the case—it would always win. " Poe also displayed his appreciation for mathematical logic in his essay "The Philosophy of Composition, " explaining that his editing of "The Raven" had followed a procedure with "the precision and rigid consequence of solving a mathematical problem. "In 2022 the AI wave produced by DALL-E's synthetic images and ChatGPT's remarkable texts attracted wide attention for its unusual, even electrifying results. In just one year, AI programs had built an audience of a hundred million weekly users (New York Times, December 7, 2023, B5). To avoid the additional investment of time and effort required in creating yet another large language model (LLM) to support it, Poe. com became instead an aggregator of existing AI sites, offering access to their contents, becoming something entirely new, a second-generation AI site. In February 2023, Poe. com made public its two subscription levels, a free plan with access to older AI models, such as GPT 3/3. 5 and Claude, and a subscription plan providing access to more recent AI models, such as GPT-4 and Claude 2. Soon Poe. com was to allow users to combine these outputs and even use voice instructions like those in Siri. The rapid public acceptance of the AI wave in 2022–2023 was due in part to the wide prior acceptance of many widespread computer helps already employing machine intelligence. It was very convenient to have at hand spelling and grammar corrections (and even thesaurus alternatives), suggestions of possible results in searches, prewritten short answers in text messages, and auto-translation of foreign languages. Elsewhere, users took for granted automatic subtitles on their TV screens and detailed GPS travel instructions on their phones and car dashboards. In February 2023 D'Angelo optimistically claimed in opening Poe. com to public use, "I expect this AI technology wave to be far more transformative than the personal computer, the internet, or the smartphone, and I expect it to have a larger and more positive impact on society and the world economy" (quorablog. quora. com/Poe-1). D'Angelo also described Poe. com as the fastest and easiest way to talk to AI, even across platforms. Thus Poe. com could access both ChatGPT from OpenAI and Claude from Anthropic, produce expandable Wikipedia-style links, and also function independently of Quora while using its informational resources. During 2023, the New Yorker and the Atlantic, prominent general magazines not noted for specializing in technology, published special AI issues. However, many magazines and newspapers were becoming deeply concerned that AI sites were borrowing their published information without payment or credit. The Associated Press and publications such as the New York Times and the Wall Street Journal entered negotiations with OpenAI for payments for such use; Bloomberg, which had extensive business news, and the Washington Post were not doing so, perhaps because they intended to pursue their own AI efforts (New York Times, December 30, 2023, B1, 3). In addition, the New York Times filed a lawsuit accusing OpenAI's ChatGPT and Microsoft's Bing Chat of copyright infringement, contending that each had produced content nearly identical to Times articles, giving them a "free-ride on The Times's massive investment in its journalism by using it to build substitutive products without permission or payment" (New York Times, December 28, 2023, B1). The conflict in AI between long-term altruistic research and immediate profitability came to a head late in 2023. The OpenAI charter had defined its purpose as "acting in the best interests of humanity, " but the forces of commercialization proved hard to fend off. For violating the altruistic precepts of its charter, the board of OpenAI fired its leader Sam Altman in November 2023. But after only a few days, in the face of near-universal threats of resignation from the staff and his decamping to a post at Microsoft, the board relented and rehired him. An unexpected consequence was that Microsoft, a late player in AI, made a multibillion-dollar investment in OpenAI. In his capacity as chief executive back at OpenAI, Altman predicted early in 2024 that online chatbots would "take a leap forward that no one expected" (New York Times, January 9, 2024, B1). He surveyed several areas of possible progress: the execution of tasks that were predictable or routine, the imitation of human reasoning, applications for robots and domestic work, the mastery of complex procedures in math and science, and success in achieving multimodal relationships among the several media (i. e. , text, photographs, voice, video, diagrams, spreadsheets, and computer programming). He warned that errors were still to be expected since statistical probability would continue to define apparent relationships; moreover, he warned that the growth of AI could take away certain jobs. Facing the widespread impact of AI, some major corporations and organizations became very uncomfortable that it was a black box that no one fully understood. Moreover, as generative forms of artificial intelligence (GAI) began to be used in more programs and procedures, issues of trust and governance grew in importance. A growing concern was that GAI might create runaway software that led to unimagined dangers, even the universal destruction of biological or nuclear warfare that ended all civilization. However unlikely, it seemed conceivable that a rogue program once in the system could be copied, recopied, and distributed with no way of stopping or even tracing it. The fear was not entirely new; images of computers taking over everything and producing a catastrophe were a staple of dystopian science fiction, memorably realized decades ago in the film 2001 (1968). One proposed alternative to the unregulated growth of propriety AI software with its possible dangers was the use of open-source software without proprietary barriers that anyone could use legally. It was recalled that several decades ago IBM had achieved more prominence with its open PC system that tolerated or even invited compatible clones than Apple had with its closed operating system. Two companies that followed the open-source approach were Meta (Facebook) and Mistral, a French company with a multibillion-dollar investment. Proponents of open source argued that multiple users could best review the code and remove errors and other undesirable elements. Opponents of open source argued that without proprietary guardrails, malicious players could more readily spready harmful agents or misinformation (New York Times, December 11, 2023, B2). The radical growth of AI applications became a serious concern, and several attempts were made to regulate its extremes. On October 30, 2023, U. S. president Joe Biden, recognizing that it was too early for detailed regulations, issued an executive order "On the Safe, Secure and Trustworthy Development and Use of Artificial Intelligence. " It recognized the possibility that AI could introduce social harms such as "fraud, discrimination, bias, and disinformation; displace and disempower workers; stifle competition; and pose risks to national security. " In international regulations, the Bletchley Declaration (named for the site where British intelligence cracked the German enigma code in World War II) signed in November 2023 by twenty-eight countries, including the United States, the United Kingdom, China, six EU member states, Brazil, Nigeria, Israel, and Saudi Arabia, resulted in the creation of the first summit on the regulation of AI (www. reuters. com/technology). As the U. S. presidential elections neared in 2024, fears arose of increased political misuse of AI to create misinformation, deep fakes, and even outright lies. The magazine Wired was alarmed that Copilot, Microsoft's AI chatbot, had already responded to questions about elections with misinformation and fake scandals—even though presumably not for political purposes (Wired online, December 15, 2023). When asked about locations for voting in the future 2024 U. S. elections, Copilot mistakenly provided an unrelated article about Vladimir Putin, and when the program pursued details of a "corruption allegation" against a Swiss lawmaker, the entire affair, as it later revealed, was an AI error or "hallucination" (New York Times, December 15, 2023, A1). In the same report, Wired revealed that Copilot's error level had varied by language; reports in English were said to be 52 percent accurate, in German 28 percent accurate, in French 19 percent accurate. Moreover, Copilot sometimes merged different packets of data, creating "something totally incorrect out of initially accurate data. " On the whole, Wired concluded, Copilot's election reports were alarmingly "systemic" as misinformation, making Copilot at that time "an unreliable source of information for voters. " Wired quoted the warning of Josh A. Goldstein of the Georgetown University Center for Security and Emerging Technology that misinformation from chatbots in reporting where and how to vote could seriously "hinder democratic processes" (www. wired. com/tag/elections/). According to a study by Vectara, AI systems varied in reliability. Thus, the error rate of AI systems in producing the misstatements called "hallucinations" varied substantially from system to system: "OpenAI's technologies had the lowest rate, around 3 percent. Systems from Meta, which owns Facebook and Instagram, hovered around 5 percent. The Claude 2 system offered by Anthropic, an OpenAI rival also based in San Francisco, topped 8 percent. A Google system, Palm chat, had the highest rate at 27 percent" (New York Times, November 7, 2023, B1). Some public figures did not understand that AI as a generative text agent could create convincing-looking but false documents. One such fake document, produced by Google Bard, was submitted by former Trump fixer Michael D. Cohen to his embarrassment in a legal proceeding regarding his court supervision (New York Times, December 30, 2023, A1). Taking a broad view of all AI issues, U. S. Supreme Court chief justice John G. Roberts Jr. warned that "any use of AI requires caution and humility. " Submitting legal documents that cited nonexistent cases, he remarked with great understatement, was "always a bad idea. " Roberts expected that in 2024 AI would play an increasingly prominent role in the legal system, one that might be either positive or negative. Roberts was alarmed by reports of law professors that "AI apparently can earn B's on law school assignments and even pass the bar exam, " but he also was optimistic that "AI could streamline legal filings and even save money, " emphasizing however that human judgment could never be replaced (New York Times, January 1, 2024, A1). The growth of AI was also an acute concern in education. For some schools and colleges from its beginnings, AI has always been subject to limitations imposed by its data collection procedures. The indiscriminate collection of everything that could be found on the Internet required little or no supervision, but it also lacked the means to filter out the unacceptable elements, such as deliberate lies, hate speech, and conspiracy theories. It remains the case that the quality of AI will always be limited by the quality of its data collection procedures. For old computer users the abbreviation for this problem was gigo, garbage in, garbage out. An unfortunate development is that AI is improving its ability to create deep fakes with images and voices that can maliciously imitate known individuals, making them apparently do such things as recite Hitler's speeches or appear in public without any clothes on (New York Times, January 8, 2024, B1). On a more mundane level, AI-produced fake reviews of products and services have become a billion-dollar industry, deceiving millions of customers every year. The initial appearance of a solid block of five-star reviews is encouraging—until it is realized that it is statistically improbable. It is sometimes difficult to decide whether a review is sincere or malicious. Indeed, rewards such as discounts, cash back, and loyalty points are used to encourage ordinary users to write and sign fake rave reviews for posting on Facebook, X (formerly Twitter), and other social media. Most cases of fake reviews are petty, but in one extreme instance orthopedist Dr. Mark Mohrmann and his practice, Highline Orthopedics, were convicted and fined 100, 000. New York AG Letitia James stated that "he and his wife worked together to suppress negative reviews and artificially inflated positive reviews of his practice on numerous websites, including ZocDoc, Google, Yelp, Healthgrades, Vitals, Md. com, RateMds. com, and the Better Business Bureau" (https: //ag. ny. gov, October 30, 2023). One unwanted effect of AI is that it removes direct contact between the makers and the users of information. In a sense, corporations do the same thing by putting distance between the producers and the consumers of goods and services, leaving investors and managers to be concerned mainly with profits and stock market values. In this connection, Ezra Klein remarked in a New York Times editorial, "Capitalism is a kind of artificial intelligence, and it's far further along than anything the computer scientists have yet coded. " The great danger of AI is the possible lack of human control, Klein warned: "There is no off switch" (November 28, 2023). In the future, AI will surely expand and deepen in importance in virtually all areas, including but not limited to education, health care, finance, manufacturing, business, science, media and communications, arts and entertainment, technology, politics, and even warfare. Already social media contain a great many promotions for AI training, applications, and assistance of all kinds. In the future, sub-miniature AI devices may become common; one example is the wearable Humane lapel pin AI device operated by voice and finger movements (www. reuters. com, November 9, 2023). Perhaps AI will have a place in managing hyperintensive data collections that are expanding in very large and very small subject areas, such as astronomy (How do galaxies form? What is antimatter? ) and microbiology (How do plant and animal cells communicate? What is consciousness? ). The October 2023 issue of Wired magazine contained six sophisticated articles on the possible dangers of AI in the next ten years when GAI produces more sophisticated software that can manufacture voices and images that are difficult for the public to detect. Although watermarking has been proposed to identify AI products, legal regulations of technology have been traditionally slow to be adopted and then swift to be insufficiently appropriate or precise. Arguing that currently neither the government nor big tech is doing enough to deal with the future problems of AI, including potentially damaging or destructive events or even a civilization-ending singularity, the editors of Wired communicated their extraordinary worry in a startling cover headline that pushed editorial propriety to its absolute limit: "Dear AI Overloads: Don't F*ck This Up. " The same warning might be extended to AI underlings and AI middlemen. This might be the place to return to the subject of Poe sites on the Internet, where we began. Eapoe. org is the irreplaceable site of the Poe Society of Baltimore; Poe. org belongs to the singer Anne Danielewski; and the site with the signature that Poe preferred in his published works, Edgarapoe. com, belongs to the Poe Society of Prague, now hibernating. Furthermore, the expression Poe's law, named for the real Nathan Poe, describes the failure on the Internet of the rhetoric of extreme parody or satire to be understood without some "clear indicator of the author's intent" (Wikipedia). In any event, if you'd like your own poe site, look into Eapoe. com, which when I wrote this was on sale for the bargain price of 2, 495. And, of course, eapoe. info is where Poe in Cyberspace columns are archived.