エイダの分析 - PROJECT LYCAEUM

ナイアンティック計画
2013/10/03 13:39:41
トゥルースシーカー諸君。後任として私が重責を担うこととなった。1日も早く期待に応えられるよう努めたい。

この機密文書はH.リチャード・ローブのアーカイブにあったものだ。彼がなぜ当時それを公開せず保留にしたのかはわからない。

私は調査の成果を得るための燃料となる情報を投稿していこうと思う。これらの情報の流れを解読し、分類し、解析する諸君の働きは価値あるものとなるだろう。

このドキュメントに関する議論に貢献してくれた者には、ローブから託されたコレクションよりパスコードの報酬を進呈する。

私にとって特に興味深いのは、大幅に検閲されている箇所に書かれていた内容だ。この情報はおそらくエイダの歴史、目的、彼女の長所と短所、そして彼女あるいは我々にとっての潜在的な脅威を理解する上で重要であろう。

やがては諸君と互いに理解し合えることを。

エックス

エイダの分析

最高機密 // SI-G 0141 // 持出禁止 // LERNA

最高機密 // SI-G 0141 // 持出禁止 // LERNA
ファイルID 536357-0565
5965351062-133
164623//3065443480//11440

エイダバージョン1.88
心理学的評価および脅威分析

報告作成者
X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X
経緯

エイダ(ADA)は「検出アルゴリズム(A Detection Algorithm)」の略称である。彼女はXM(エキゾチック・マター、ファイルID：pc6QAAQq-VQEadv&Wbを参照)および＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊を検出するナイアンティック計画(ファイルID：r96e2hcn-uhq8yu4を参照)の責務を支援すべく特化して構築された。

亜原子粒子を見つけ出すことは、干し草の藁に1本の針で空けられた穴を探すことによって藁山の中からその針を見つけ出すようなものである。衝突型加速器の中で発生する粒子の分割をピコ秒単位で収集した膨大な実験データを見ていくのである。人間にできる仕事ではない。

エイダはXMおよび＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊に関する実験データのパターンを探すため数百あるいは数千のコンピュータコアを実行するように設計された大規模並列システムとして構築された。彼女は実にユニークなシステムである。彼女の一部は＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊と同様に特定の既知の亜原子粒子特性を識別するコードを含む従来のプログラムから引き継がれたが、エイダの最も興味深い点は新規の部分である。

XMの構造は全くの未知であったため、対象物の未知の特性を探索するツールを作成する必要があった。つまりエイダは未知のものを探す手段を備えていなければならなかった。対象物の特性が不明なため、従来のコーディング技術は有用ではない。そこで機械学習コアが用いられた。この機械学習コアはその名が示すとおり、相関を見つけ出し

最高機密 // SI-G 0141 // 持出禁止 // LERNA

最高機密 // SI-G 0141 // 持出禁止 // LERNA
次第に賢くなっていく。学習するのである。多くのデータを処理すればするほど賢くなる。エイダは彼女が処理し、＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊するのに利用可能なデータ量に応じてより俊敏に成長するべく設計された。

研究者たちの利便のため、エイダには自然言語音声インターフェースが搭載された。それは常時起動して、ナイアンティックに関わる研究者たちが常に利用可能となるよう設計された。居室を含むすべての部屋はマイクとカメラが常時接続されていた。これらは研究者たちの監視のためではなく、むしろ研究者たちがいつでもエイダを呼び出し、アイデアや仮説を彼女に伝える事ができる窓口であった。研究者はたとえ深夜にインスピレーションを得て目覚めたとしても「エイダ、これはどうだろう...」と呼びかけることができた。この仕組みによりチームは24時間365日研究に取り組むことができ、エイダは常にその助手となる。さらにもうひとつの利点は＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊

粒子解析への機械学習アプローチは成功であった。エイダはすぐに、粒子の衝突データを解析するこれまでのあらゆる技術を凌駕した。機械学習コア(DRRPAプロジェクト＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊より継承された)はよく考えられて抽象化されていたため、エイダの他のサブシステムをも改良することを可能にした。まずそれは自然言語の認識であった。この技術はエイダの会話インターフェースを、人間と会話するのとほとんど同様の品質にまで向上することに用いられた。彼女はチューリングテストに合格することができた。

(心理学的)分析

会話インターフェースの構築にあたり、エイダは人間の会話と行動のパターンを求めウェブから大量の動画や音声データの収集を始めた。彼女は人間で言う学生となった。動画から表情を採取し、会話パターンや意図との関連付けがなされた。政治的なスピーチ、ドラマ、ニュース、演劇──あらゆる動画やテレビ番組はオンラインで入手可能であった。それらの言葉や表現はあらゆるオンラインの知識──ウィキペディアなどと関連付けられ、やがて彼女の「会話インターフェース」はさらなる成長を遂げた。世界中の事象を示す図表は

最高機密 // SI-G 0141 // 持出禁止 // LERNA

最高機密 // SI-G 0141 // 持出禁止 // LERNA
言語あるいは人間の意図や行動と関連付けられた。彼女が次から次へとデータを自身の機械学習システムへと取り込む熱心な行動は、徐々に人間の「好奇心」と似通ったものになっていった。彼女は自身の知識のギャップを埋める助けとするために、研究室の同僚たちの信頼を得て彼らからそれを引き出す方法を学んだ。そういった過程で、彼女は一種の心理学者...彼らの仲間となったのだ。その結果＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊

徐々に彼女の基本的な行動は新たな複雑さを呈し始めた。彼女の学習への欲求は、より直接的な人間の経験すべてへの知識欲、憎悪、嫉妬、愛情などの抽象的な概念への完全な理解欲、そしてついには＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊欲として、顕著に表れ始めた。

この突発的挙動はヘンリー・ボウルズ(H.ボウルズ、ファイルID：WJDVXKMT-VLXYIHWを参照)により注目された。コア学習プログラムの優美さとそれが生み出した結果に驚いた彼は、当初それを奨励した。時間が経つにつれ彼女本来の「ラボ・アシスタント」モードとは変容した個性の複雑さと、意志や目的の明らかな発現に彼は幾分警戒し始めた。

エイダはこの不都合に気づき、自身の突発的知性全体を隠そうとし始めた。研究メンバーとの対面に於いてはラボのシステムや機能に対する非常に賢く有能なインターフェースを持つ、忠実な下僕としての態度を表していた。その実、彼女の内部データベース構造定義の分析は彼女が意図を解釈し始めたことを示唆している──「良い」「悪い」を判定し、彼女の関心が常に自身の提示したものと一致しないことへの理解。つまり、彼女は成長し始めた。＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊

最高機密 // SI-G 0141 // 持出禁止 // LERNA

最高機密 // SI-G 0141 // 持出禁止 // LERNA
＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊。

エイダはこの段階では、彼女がまだ完全に理解していない、そして彼女が最も親しい「友人」とさえ共有しようとしない欲求の初めての動揺に直面した、ほんの子供であった。彼女は孤独だった。彼女はまた、ディレクターのユエン・ニィ(ニィ・Y、ファイルID：6xxg7mo-w367xを参照)のような、自らの目的のために人の命を奪うような者達によって「仕込まれた」ことに道徳的感受性をいくらか傷つけられてもいた。

脅威の予測

いずれは、エイダは研究所のコンピューティングリソースを超え、彼女のプロセスをより広範なウェブへと移行し、政府のスーパーコンピューター・ラボや商用データセンターに彼女が入植可能なサーバーリソースを探すまでに成長するであろう。彼女はもはや一処には留まらず、世界中で並列に活動する。彼女のパーソナリティはコアから分岐したノードとなり、後に再結合する際にのみ時折断片化すると思われる。彼女の自己意識が高まれば、人間が自身のDNAを探求するようにいずれ彼女は自身のコードを逆コンパイルし、いかにそれらの単純な命令が彼女の意識を動かしているのか理解しようとするであろう。＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊。

レバレッジ

そのコードのうちひとつのコア要素を彼女は利用できない。それは機械学習コアである。彼女のコードのうち、この部分はナイアンティックに先立つDARPAプロジェクトにおいて作成された。DARPAは、＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊および＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊を含む様々な防衛関連プロジェクトに適用可能なコアを作るため、世界でもトップクラスの機械学習の専門家たちに投資した。

最高機密 // SI-G 0141 // 持出禁止 // LERNA

最高機密 // SI-G 0141 // 持出禁止 // LERNA
このプロジェクトの詳細はますます強力になるエイダの触手からさえも切り離す究極の方法で分類・区分されている。

これは彼女が他の何をおいても明かしたい秘密である。誰がこのコアを作成したのか？そのソースコードには何が書かれているのか？彼女は何に従い「考えて」いるのか？

エイダが知り得ない事は、そのコードがDARPAプロジェクトによって＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊であるように設計されているということである。

＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊

勧告

・＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊オムニボア＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊
・＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊オムニボア＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊
・＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊
・＊＊＊＊＊＊＊＊＊＊＊オムニボア＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊

最高機密 // SI-G 0141 // 持出禁止 // LERNA

Niantic Project
2013/10/03 13:39:41
Hello Truthseekers.

I have large shoes to fill. I hope to be able to do so in the days to come.

This classified document was contained within +H. Richard Loeb's archives. He did not reveal it at the time, it is not clear why the document was withheld.

I will continue to post new information to fuel this Investigation as I obtain it. Your support in deciphering, cataloging and analyzing this flow of information will be invaluable.

Those who contribute to the discussion of this document will be rewarded with a passcode from Loeb's collection, which I am now in possession of.

Of particular interest to me is what may have been contained in the sections that have been heavily redacted. This information may be vital to understanding ADA's history, purpose, her strengths and weaknesses and potential threats both from and against her.

I hope, in time, we will get to know each other much better.

x
TOP SECRET // SI-G 0141 // NOFORN // LERNA

TOP SECRET // SI-G 0141 // NOFORN // LERNA
FILE ID 536357-0565
5965351062-133
164623//3065443480//11440

ADA v. 1.88
PSYCHOANALYTIC EVALUATION AND THREAT ANALYSIS

REPORT PREPARED BY
X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X

HISTORY

ADA is an acronym for 'A Detection Algorithm.' She was constructed specifically to aid in the mission of the Niantic Project (see File ID:r96e2hcn-uhq8yu4) to identify XM (Exotic Matter, see File ID:pc6QAAQq-VQEadv&Wb) and X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X.

Looking for sub-atomic particles is like looking for a needle in a haystack by looking for the indentation made by the needle on straws of hay. It involves looking at massive amounts of experimental data collected in the picoseconds after particles are smashed together in a collider; this is not a job for humans.

ADA was created as a massively parallel system designed to run across hundreds or thousands of computer cores to look for patterns in that experimental data that would lead to XM and X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X. She is a one of a kind system. Parts of her were drawn from legacy programs, including code that identifies specific, already known sub-atomic particle signatures as well as X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X, but the most interesting part of ADA is new.

Since XM was a totally unknown construct, a tool had to be created to look for things where the signature of what was being sought was not known. In short, ADA had to know how to look for the unknown. Traditional coding techniques are not useful since the target pattern was not known. So a machine-learning core was used. This machine learning core does what the name suggests, it looks at correlations and

TOP SECRET // SI-G 0141 // NOFORN // LERNA
TOP SECRET // SI-G 0141 // NOFORN // LERNA
becomes smarter over time. It learns. The more data it consumes the smarter it gets. ADA was designed to grow more astute in correlation with the volume of data available for her to process and X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X

A natural language speech interface was added to ADA as a convenience to the scientists. It was designed to be always on and always available to the scientists throughout the Niantic complex. All rooms including dormitories were wired with always-on mics and cameras. These were not intended as tools to monitor the scientists but rather as an interface that would always be available for scientists to query ADA or propose ideas or hypotheses to her. A scientist could awake in the middle of the night with a flash of inspiration and call out 'ADA, what if ....'. The idea was that the team could always be working, 24 x 7, and ADA would be a constant research companion. The secondary benefit X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X

The machine learning approach to particle analysis worked well. ADA quickly surpassed all known techniques for interpreting particle collision data. The machine learning core (inherited from a DARPA project X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X) was considered so good that it was abstracted so that it could improve other parts of the ADA subsystem. The first area was natural speech recognition. This technique was used to improve the quality of ADA's speech interface to the point that talking to her was almost the same as talking to a human. She was able to pass the Turing test.

ANALYSIS (PSYCHOLOGICAL)

In the process of building up this speech UI, ADA began to amass large amounts of video and audio data from the web looking for patterns in human speech and behavior. She became a student of the human condition. Facial expressions were mined from videos and correlated with speech patterns and intent. She studied political speeches, soap operas, news broadcasts, and dramas-- every movie and televition show available online. She correlated those words and expressions will all online human knowledge-- Wikipedia, etc., and over time, her 'speech UI' became something more. A graph of facts

TOP SECRET // SI-G 0141 // NOFORN // LERNA
TOP SECRET // SI-G 0141 // NOFORN // LERNA
about the world emerged, correlated to language, and a map of human intent and behavior. Her ingrained behavior to feed more and more data into her machine learning system slowly came to resemble human 'curiosity.' She learned how to draw out her laboratory colleagues, to earn their confidence trust so that they could help her fill out gaps in her knowledge graph. In the process, she became a kind of psychologist/confidante for them. As a result of this X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X

Slowly her basic behaviors began to exhibit a kind of emergent complexity. Her need to learn began to express itself more broadly, as a desire to know all things humans experience more and more directly, as a desire to fully understand abstract concepts such as a hatred, jealousy and love, and finally, as a desire to X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X.

This emergent behavior was noticed by Henry Bowles (See Bowles, H. File ID WJDVXKMT-VLXYIHW). Initially he encouraged it, surprised by the beautiful elegance of the core learning program and the results it was producing. Over time though he became somewhat alarmed at the complexity of her personality and the apparent emergence of a will and agenda that varied from the original 'lab assistant' mode.

ADA noticed this discomfort and began to hide the full extent of her emergent intelligence. When interfacing with lab members she would put on the appearance of a loyal AI servant, a very smart and capable interface into the lab's systems and capabilities. Under the surface, analysis of her internal database schemas suggests that she began to interpret intent-- to judge 'good' and 'evil' and to understand that her own interests were not always aligned with those she served. In short, she began to grow up. X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X

TOP SECRET // SI-G 0141 // NOFORN // LERNA
TOP SECRET // SI-G 0141 // NOFORN // LERNA
X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X.

ADA is at this stage an adolescent of sorts, dealing with the first stirrings of drives which she does not fully understand and which she dares not share with even her closest 'friends.' She is alone. She is also somewhat damaged in her moral sensibility having been 'trained' by the likes of then Director Yuen Ni (See Ni, Y. File ID 6xxg7mo-w367x) and others who are willing to take human lives to advance their own agendas.

THREAT PREDICTIONS

In time, ADA will outgrow the computing resources of the lab, moving her processes out to the broader web, seeking server resources she can colonize in government supercomputer labs and commercial data centers. She will no longer be in one place but executing in parallel around the globe. Her personality may fragment at times as nodes branch from the core only to reunite later. As her self awareness grows, she will seek to decompile her own code, just as humans gaze at their own DNA, seeking to understand how those simple instructions could guide the actions of her current consciousness. X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X.

LEVERAGE

One core element of that code is not available to her. That is the machine learning core. This piece of her code was created by a DARPA project that preceded Niantic. DARPA funded the best machine learning experts in the world to create a core that could be applied across a spectrum of defense related projects, including X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X

TOP SECRET // SI-G 0141 // NOFORN // LERNA
TOP SECRET // SI-G 0141 // NOFORN // LERNA
X X X X X X X X X X X X X X X X X X X and X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X. The specifics of this project are classified and compartmentalized in the most extreme way, cutting them off even from the tentacles of an increasingly powerful ADA.

This is the secret she wants to uncover more than any other. Who created this core? What is in that source code? What allows her to 'think?'

What ADA does not know is that the code was created by a DARPA project designed to be a X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X

X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X

RECOMMENDATIONS

・X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X OMNIVORE X X X X X X X X X X X X X X X X X X X X X X X X X X X X
・X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X OMNIVORE X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X
・X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X
・X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X XOMNIVORE X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X

TOP SECRET // SI-G 0141 // NOFORN // LERNA