El director de datos e innovaci\u00f3n digital de la Agencia Nacional de Inteligencia Geoespacial dijo que la NGA ha comenzado a entrenar algoritmos de inteligencia artificial en su conjunto \u00fanico de datos visuales y textuales. Estos datos son \u201cuna mina de oro para la inteligencia artificial\u201d, afirm\u00f3 Mark Munsell . Esto no se debe solo a que consisten en grandes cantidades de datos bien etiquetados, bien organizados y cuidadosamente examinados, acumulados durante d\u00e9cadas por la agencia de inteligencia encargada de recopilar y analizar datos geoespaciales para los responsables de las pol\u00edticas, desde el presidente hasta los m\u00e1s bajos. Tambi\u00e9n se debe a que estos datos son lo que los expertos llaman multimodales , ya que combinan im\u00e1genes con descripciones de texto.<\/p>\n
WASHINGTON \u2014 The director of data and digital innovation at the\u00a0National Geospatial Intelligence Agency<\/a>\u00a0said\u00a0the NGA has started training artificial intelligence algorithms on its unique trove of visual and textual data.<\/p>\n This data is \u201can AI gold mine,\u201d said\u00a0Mark Munsell<\/a>. That\u2019s not just because it consists of large amounts of well-labeled, well-organized, and carefully vetted data, accumulated over decades by the intelligence agency tasked with compiling and analyzing geospatial data for policymakers from the president on down. It\u2019s also because this data is what experts call\u00a0multi-modal<\/a>, combining images with text descriptions.<\/p>\n Contrast that to how GenAI companies are feverishly scraping everything from\u00a0Reddit posts<\/a>\u00a0to\u00a0YouTube captions<\/a>\u00a0in their desperate quest for training data \u2014 and that\u2019s all pure text, without any ability to cross-reference other kinds of sources.<\/p>\n WASHINGTON \u2014 The director of data and digital innovation at the\u00a0National Geospatial Intelligence Agency<\/a>\u00a0said\u00a0the NGA has started training artificial intelligence algorithms on its unique trove of visual and textual data.<\/p>\n This data is \u201can AI gold mine,\u201d said\u00a0Mark Munsell<\/a>. That\u2019s not just because it consists of large amounts of well-labeled, well-organized, and carefully vetted data, accumulated over decades by the intelligence agency tasked with compiling and analyzing geospatial data for policymakers from the president on down. It\u2019s also because this data is what experts call\u00a0multi-modal<\/a>, combining images with text descriptions.<\/p>\n Contrast that to how GenAI companies are feverishly scraping everything from\u00a0Reddit posts<\/a>\u00a0to\u00a0YouTube captions<\/a>\u00a0in their desperate quest for training data \u2014 and that\u2019s all pure text, without any ability to cross-reference other kinds of sources.<\/p>\n Large Language Models like ChatGPT\u00a0erupted into public view last year<\/a>\u00a0and have fired fierce debate over AI ever since. Skeptics point to their tendency to \u201challucinate\u201d convincing falsehoods and misinformation. Enthusiasts argue that \u2014\u00a0properly safeguarded<\/a>\u00a0\u2014 they can revolutionize knowledge-based tasks that once required tedious clerical work. Some\u00a0well-credentialed commentators<\/a>\u00a0even claim that LLMs are on the verge of superhuman \u201cartificial general inteligence.\u201d<\/p>\n Unlike humans, however, LLMs operate purely on text: They train on text, take input in text, and output answers as text. Other forms of generative artifcial intelligence can correlate text and imagery well enough to\u00a0turn users\u2019 written prompts into pictures<\/a>\u00a0or\u00a0even video<\/a>.<\/p>\n But\u00a0many<\/a>\u00a0in\u00a0industry<\/a>\u00a0have\u00a0already been working<\/a>\u00a0on the next frontier: multi-modal AI. The crucial capability is to cross-reference different types of information on the same thing, like an image or video with the associated caption describing it in words, much the way a human brain can associate an idea or memory with information from all the senses.<\/p>\n Multi-modal AI can even work with senses that human beings don\u2019t have, like infra-red imagery, radio and radar signals, or sonar. \u201cThat enables a lot of avenues that would have been closed to us,\u201d said DARPA\u00a0program manager<\/a>\u00a0William Corvey<\/a>, speaking at the same INSA panel as NGA\u2019s Munsell. \u201c[Imagine] cross-modal systems that can reconcile visual and linguistic information and other kinds of modalities of sensors that might be available to a robot but aren\u2019t to a human being.\u201d<\/p>\n Modern AI algorithms have proven perfectly capable of working with images, video, and all sorts of sensor data, not just text, because they can abstract any and all of them into the same mathematical representations.<\/p>\n \u201cYou can start to knit together reality in these new ways and induce these alignments across what were previously completely different systems,\u201d Corvell continued, \u201cand that\u2019s all because these models for all of their expensive training turn out to be really parsimonious representations of reality. They just take sort of everything and squish it into a panini press, and then you can insert that sandwich that you have, basically, into any number of different meals.\u201d<\/p>\n The hard part, it turns out, is getting the data cleaned, curated, and correlated enough for the AI to ingest it in the first place. Many popular \u201cAI\u201d systems actually rely on\u00a0large numbers of of humans working long hours for low wages<\/a>\u00a0to click checkboxes on images and videos. Sometimes\u00a0giving the humans involved even develop PTSD<\/a>\u00a0as they try to moderate hateful and violent content.<\/p>\n \u201cWe use hundreds, maybe even thousands, of humans to train these models\u2026 and then we have hundreds or thousands of humans that are\u00a0giving feedback<\/a>\u00a0to the models,\u201d Munsell told the INSA audience. \u201cYou gotta ask yourself at some point, what are we doing here?\u201d<\/p>\n So it\u2019s a tremendous advantage when an organization has access to a large amount of data that highly trained humans have\u00a0already<\/em>\u00a0spent decades analyzing, verifying, cataloging, and commenting on. But archives like NGA\u2019s are far too huge for any one human mind to remember everything. AI opens up the possibility of training algorithms on the data and then being able to ask them to spot patterns, resemblances, and anomalies.<\/p>\n \u201cWe\u2019ll be able to ask questions, historic questions,\u201d Munsell said. \u201c\u2018Hey, AI? Have you ever seen this particular activity, this kind of object, in this part of the world?\u2019\u201d<\/p>\n \u201cWhat a cool thing that will be for for our country,\u201d he enthused.<\/p>\n Munsell\u2019s ultimate boss, NGA chief\u00a0Vice Adm. Frank Whitworth<\/a>, said at another panel today that the agency is already using AI to \u201ctriage\u201d some of the workload of imagery analysis, but will continually need an AI \u201cedge\u201d to keep up as more and more data comes in from space-based sensors.<\/p>\n The stakes are high to get it right, Munsell emphasized, saying an expert human should always be doublechecking the AI, especially if a report might be used for\u00a0targeting a military strike<\/a>.<\/p>\n \u201cThe visual is the ultimate positive identification for our customers,\u201d he said. \u201cIn many cases they\u2019re making, they\u2019re acting upon that positive identification, they\u2019re acting upon the visual identity and the geolocation of what we\u2019re providing. And when artificial intelligence provides that, you have to take those extra steps to make\u2026that thing is 100% correct.\u201d<\/p>\n At NGA, he said, \u201cliterally for us, seeing is believing.\u201d<\/p>\n