
ここからコンテンツです。
Shigeki Nakauchi
Today we are living in an age of artificial intelligence (AI), where it is feared that someday soon computers may threaten our jobs as they gradually surpass human ability. But is technological singularity really something to worry about? Prof. Shigeki Nakauchi says that the key to this answer lies in the differences between AI and humans. Prof. Nakauchi, who is an expert in the function of the brain that controls visual perception in humans and its mechanism, explains that, "while they differ from one another in that one is made up of protein and the other of semiconductors, brains and computers are similar in that they are both made up of substances, and there is no guarantee that there are things only humans can do. However, there are at least big differences in what humans and computers can do at present." Prof. Nakauchi is trying to gain a glimpse into a future where humans and AI may coexist by addressing the fundamental question of what it means to be human.
Interview and report by Madoka Tainaka
Prof. Nakauchi became interested in computers when he was a middle school student and the computer boom was just beginning. He entered a technical college (KOSEN) hoping to research information engineering, but eventually turned his attention to biology. This was because he was driven by an interest in testing the limitations of machines and finding out how they differ from humans as living beings. For Prof. Nakauchi the key to these questions was to focus on learning about human beings.
Prof. Nakauchi is still inspired by these questions. The current debate in this field focuses on the possibilities thrown up by the third AI boom that the singularity (a hypothetical technological singularity when AI supersedes the human brain) may be within reach. In this context, Prof. Nakauchi believes that researchers should be focusing on the differences between machines and humans and in particular how machines and humans may communicate with each other.
"Computers are able to compute much more data than the human brain at much faster speeds, but generic AIs that supersede human ability are not likely to be developed for some time. On the other hand, humans are gradually adapting to a society that utilizes a plethora of machines. As one example, the Internet now serves as an external memory device for humans, meaning that we no longer need to remember maps or unfamiliar words. This causes the brain and society to change with it.
The trend towards ever greater human-machine coexistence appears unstoppable. In order to prepare for such a future, my goal is to uncover how to achieve optimal human-machine relations by conducting an in-depth study of humans."
Broadly speaking, Prof. Nakauchi's research can be divided into two categories. The first is the basic groundwork of "scientifically studying visual perception" with the aim of uncovering visual function. The second involves consolidating the findings obtained through this basic research and seeking technological applications, in other words "applying visual perception as technology".
Recently, in particular, this basic research has produced a string of ground-breaking discoveries, the results of which have been published in international journals1. One such study, relating to "distinguishing between textures" using visual perception, uncovered what humans use to distinguish between "reflective materials" such as mirrors and polished metals which produce a reflective effect through light reflecting on the surface and "transparent materials" such as glass and ice through which light passes and refracts.
According to Prof. Nakauchi, "humans have good eyes and can discriminate slight differences between various materials. For example, humans can tell that a desk which looks like wood is actually made of plywood, or whether or not leather is real. The same applies to glass and metals, with humans being able to distinguish between these materials even if they have complicated shapes. However, light reflects only on the surface of metal materials and is partially absorbed by and scattered in glass materials, meaning that these materials have no color of their own and merely distort and reflect whatever is around them. Up until recently, it wasn't clear what humans used to distinguish between the two."
Prof. Nakauchi and his team generated images using computer graphics that looked like either metal or glass and had participants say which material they thought the picture looked like. The team noted that the participants found it easier to discern the material when the object rotated rather than when they viewed a still image. Through controlling the shape and movement components of the objects to see how differently people perceived them, they achieved truly fascinating results.
"We found that, with metals, light reflection moves in the same direction as the rotation of the object, but glass reflects transmitted light and has some component in which light reflection moves in a direction opposite to rotation of the object. We also discovered that information on movement is not just used for distinguishing between materials but also gives materials that real feeling of being metallically shiny or transparent."
Interestingly enough, Prof. Nakauchi stated that humans tend to treat translucent materials such as marble or jade as important and expensive. This is why many bottles used for expensive cosmetic products are translucent.
"If we can further investigate what humans use as clues to distinguish between textures, such findings could be useful for future product development," suggests Prof. Nagauchi regarding possible outcomes of his study.
Studies concerning the face are also of great interest. People not only have the ability to determine that a person is still the same person despite a change in expression, hair style or clothing but can also determine "face-likeness" in car bonnets and paintings made up of objects, such as Arcimboldo’s portraits using fruit and vegetables. Because we need to be able to recognize faces for communication within society, humans are believed to be very good at discerning faces.
In order to investigate the visual information processing that takes place when a human looks at a face-like object, Prof. Nakauchi and his team measured brain activity and brainwaves when a subject was shown an object that looked like a face and studied the stage at which face-likeness recognition occurs2. All studies used black and white images which were pictures of human faces, as well as cars, bugs, paintings by Arcimboldo, and other objects. Participants were also shown upside-down images.
"The results of the brainwave measurement indicated that face-likeness is first processed at a very early stage of around 100ms, a face is detected at around 170ms, and expressions and individuals are identified at 250ms, meaning that facial recognition occurs in three stages. In addition, when a face is rotated, facial recognition slows to the same speed as when a face-like portrait is viewed, which tells us that processing patterns vary depending on how much an object looks like a face."
In the initial stage of the first 100ms after viewing an object the processing appears to be instinctive rather than being consciously controlled. In other words, humans almost instantaneously take note of face-likeness, even before they are aware of what the object is.
Prof. Nakauchi explains that, "We believe that face-likeness recognition is most likely an inborn function, which can be seen in how a baby needs to recognize the face of its mother to survive."
Prof. Nakauchi also studies the "science of inspiration," which is a subject unique to his team. Through studying the time at which inspiration occurs by measuring pupils, which are said to represent a person’s mental state, the team made the remarkable finding in tests that the participants’ pupils began to dilate before they were consciously aware of particular thoughts3.
"We showed participants 6-second videos of black and white dots that moved and changed density to depict various images, like a coffee cup. The video was stopped at 1.5 seconds and participants were asked whether or not they recognized the object. Then, the video was resumed and played until the end and participants were asked the same question again.
The people who answered "no" when asked the question at the 1.5 second mark provided us very interesting results. Those who answered "no" then but went on to figure out the object after watching the whole video presented pupil dilation before the one second mark. In other words, their brain had begun the process of piecing together memories even though they were not aware of this. Meanwhile, the participants who failed to recognize the object even after viewing the entire video presented little pupil dilation."
People often experience times where they can picture someone’s face but their name remains on the tip of the tongue. This kind of phenomenon may be one instance of knowing the answer but not being conscious of that knowledge.
Prof. Nakauchi says that, "applying these findings may make it possible in the future to measure how much people understand by measuring their pupils or to support memory retrieval by externally controlling pupil dilation." Future applied research is also expected to yield exciting results.
Now in his 50s, Prof. Nakauchi wants to focus on the themes that really touch his heart for the remainder of his career as a researcher. Wanting to ask the simple but big questions that have largely been neglected by science, Prof. Nakauchi now turns his attention to studying how perceptions such as preference, novelty and intimacy, which affect human behavior and decision making, are generated. "Studying aesthetics falls under this category. Why do we feel that pieces of art by famous artists are good? How do we spot a fake? It’s quite a mystery, isn’t it?
Also, since I was in a brass band as a student and am now a member of a choir for my local opera, I’m also interested in finding out the state of the brain when humans are in synch or have fostered a sense of unity when playing music or a team sport together. Figuring this out should definitely prove to be useful for coexisting with robots as well."
Evidently, Prof. Nakauchi still has a lot of interesting topics to sink his teeth into along his way to finding out what it means to be human.
人工知能(AI)が人間の能力を超え、近い将来、人々の職を奪うのではないかと危惧される昨今。はたしてそのような「シンギュラリティ」は、本当に訪れるのだろうか。その答えのカギは、「AIと人間の違いにある」と中内茂樹教授は語る。人間の視覚を司る脳機能や、そのしくみの解明を試みてきた中内教授は、「タンパク質と半導体の違いはあるにせよ、脳もコンピュータも物質から成るという点では同じであり、人間にしかできないことが存在する保証は何もない。ただ、少なくとも現状はできることに大きな違いがある」と語る。人間をより深く知ることで、AIと人間が共存する未来の姿を探ろうとしている中内教授に話を聞いた。
中内教授は、中学生の頃、当時、勃興期にあったコンピュータに惹かれ、情報工学の研究者をめざして高等専門学校に進学した。やがてその関心は生物学へ向かう。機械にできないことは何か、機械と生物(人間)は、何がどう違うのか、それを解明するためには生物をより深く知る必要がある、という思いに駆られたためだ。
その興味は現在も変わらない。むしろ、第3次AIブームを背景に、シンギュラリティ(AIが人間の脳を超えるとされる技術的特異点)がさかんに議論されるいまこそ、人間と機械の違いや両者のコミュニケーションのあり方について、科学者が方向性や可能性を示すべきだと中内教授は言う。
「膨大かつ高速の計算など、コンピュータが人間をはるかに凌ぐ領域はありますが、人間を超える汎用的なAIはまだまだ開発されそうにありません。一方で、人間は機械の溢れた社会に柔軟に適応して変わりつつある。たとえば、インターネットはいまや人間にとって外部記憶装置としての役割を担い、私たちは地図や知らない単語などをいちいち覚えておく必要はありません。それに伴い脳も変わるし、社会も変化しています。
今後ますます人間と機械の共存が不可避である以上、人間をより深く知ることで良い関係性を探り、最適なシステムを提示したいというのが私の研究のモチベーションになっているのです」
中内教授の研究は大きく分けて、視覚機能の解明をめざして「視覚を科学する」基礎研究と、基礎研究で得られた知見を技術として結晶させる「視覚を技術する」応用研究の二つに分けられる。
とくに最近、前者の基礎研究において相次いで画期的な成果を生み出し、国際ジャーナルに発表した。その一つ、視覚の「質感の見分け」に関する研究では、鏡や研磨された金属のように光が物体表面で鏡面反射する「反射物質」と、ガラスや氷のように光が物体を透過したり屈曲したりする「透過材質」について、人間が何を頼りに判別しているのかを明らかにした。
「人間の目は優れていて、この机は木目に見えるけれど合板だなとか、この皮は本物だなといった具合に、さまざまな物質の微妙な違いを見分けることができます。ガラスと金属についても同様で、それらが複雑な形をしていても判別できるでしょう。ところが、金属は表面のみで光が反射し、ガラスは光の一部が物質に吸収されて散乱するので、材質そのものには色はなく、周りの景色が歪んで写り込んで見えるだけです。これまで人間がいったい何を手掛かりにこれらを見分けているのかがわからなかったのです」と中内教授は説明する。
そこで中内教授らは、コンピュータグラフィックスで金属とガラスそれぞれに見える画像を生成し、実験参加者に金属とガラスのどちらに見えるかを聞いた。その際、静止画像よりも回転している動画のほうが見分けがつきやすい点に着目。物体の形状や動きの成分をコントロールすることで、見え方の違いを調べたところ、興味深い結果が得られた。
「じつは金属では、物体が回転する方向と同じ方向に光の反射が動くのですが、ガラスでは透過した光が反射して、物体の回転とは逆方向に動く成分も存在することがわかりました。また、そうした動きの情報は、材質の見分けに使われているだけでなく、キラキラとした金属感や透明感という材質のリアリティを生み出していることもわかったのです」
興味深いことに、中内教授が、人間は大理石や翡翠など、半透明なものを大切なもの、高級なものと感じる傾向があると教えてくれた。だから、高級化粧品には半透明なボトルが採用されていることが多いのだという。
「人間が質感を見分ける手がかかりをより詳細に解明できれば、こうした知見を将来的には商品開発などに役立てることができるかもしれません」と、中内教授は研究の出口についても示唆した。
「顔」に関する研究も興味深い。人は、表情や髪型、スタイルなどが変わっても同じ人だと判別できるだけでなく、果物や野菜などを寄せ集めて肖像画を描いたアルチンボルドの絵や、車を正面から見た場合も顔らしいと感じる。顔の認知は社会的コミュニケーションに不可欠であることから、人は顔の認知に敏感だと考えられる。
その際に行われている人間の視覚情報処理を探るために、中内教授らは「顔のように見える」物体を見たときの行動と脳波を計測し、顔らしさの認知がどの段階で生じるのかを調べた。なお、実験にはいずれもモノクロ画像で、人間の顔のほか、車や昆虫、アルチンボルドの絵などを用い、それらを逆さまに提示した場合についても調べた。
「脳波の計測の結果、まず100msという非常に早い段階で顔らしさを処理し、170msくらいのところで顔を検出、250msで表情や個人を特定するという、三段階で顔の認知を行なっていることがわかりました。また、反転させた場合、顔らしい絵ほど処理が遅れるなど、顔らしさの程度によって処理のパターンが変化することもわかりました」
100msというと意識的に処理をコントロールする前の段階である。つまり人は、物体像が意識にのぼる前にほぼ自動的に顔らしいものに注意を向けているのだ。
「赤ちゃんにとって、お母さんの顔の認識が生命をつなぐのに欠かせないように、おそらく顔らしさの認識というのは人間に生得的に備わっている機能と考えられます」と中内教授は言う。
さらに、「ひらめきを科学する」というユニークな研究も行なっている。人の精神状態を表すとされる瞳孔の変化を計測し、ひらめきがいつ生じたのかを調べたところ、実験参加者がひらめいたと意識するよりも前に瞳孔が開き始めるという、非常に興味深い知見を得た。
「提示したのは、モノクロのドット(点)が動いて、密度が変化することでコーヒーカップなどのイメージ画像を描く6秒間の動画です。その際、1.5秒でいったん動画を止め、わかったかどうかを聞き、その後、動画の続きを最後まで再生して、ふたたびわかったかどうか尋ねます。
面白かったのは、1.5秒でわからないと答えた人たちです。その時点ではわからないと答えた人も、最後まで動画を見てひらめいた人の場合は、1秒より前の段階から瞳孔が開き始めていたのです。つまり、本人は自覚していないけれど、じつは脳では記憶照合のプロセスは始まっていたことになる。一方、最後までわからなかった人の瞳孔の開きは弱いままでした」
よく、顔が浮かび、喉元まで名前が出かかっているのに、なかなか答えられないということがある。これなども、意識が届かないところでは答えがわかっているのだろう。
「この発見を応用することで、今後は瞳孔から人の理解度を測ったり、瞳孔の開きを外部からコントロールすることで、記憶の検索を支援したりすることが可能になるかもしれません」と中内教授。今後の応用研究にも大いに期待が膨らむ。
(取材・文=田井中麻都佳)
現在、50代の中内教授は、残りの研究者人生を、心の底から惹かれる、そんなテーマに打ち込みたいと語る。たとえば、嗜好性、新規性、親密さといった、人間の行動や意思決定に影響を与える感覚がどのように生じるのか、これまであまりサイエンスの俎上に載らなかったような、素朴で重大な疑問に迫る研究をしたいと言う。
「審美眼の研究もその一つ。なぜ偉大な美術家の作品をいいと思えるのか、なぜ贋作だとわかるのか、不思議ですよね。
もう一つ、私は学生時代、吹奏楽に打ち込み、現在は市民オペラで合唱団の一員として活動しているのですが、音楽の合奏やチームスポーツなどで、一体感が醸成されたり、息の合ったプレイができたりしたときの脳の状態についても探ってみたい。それがわかれば、人間とロボットの共存にも、必ず役立つはずです」と中内教授。
中内教授の人間を探る研究は、ますます面白味を増しそうだ。
Dr. Shigeki Nakauchi
Dr. Shigeki Nakuchi received his B.A., M.S., and PhD degrees in information engineering in 1988, 1990 and 1993 respectively from Toyohashi University of Technology, Aichi, Japan. He was a visiting researcher at Lappeenranta University of Technology in Finland from 1998 to 1999. Since 2002 and 2007, he has been a visiting researcher and a visiting professor at Brain Science Institute in RIKEN and Kochi University of Technology respectively. He is currently a professor at the Graduate School of Computer Science and Engineering Department, Toyohashi University of Technology. He is a docent at the University of Eastern Finland in the field of Color Vision Science and Technology as from 1 November in 2017.
Madoka Tainaka is a freelance editor, writer and interpreter. She graduated in Law from Chuo University, Japan. She served as a chief editor of "Nature Interface" magazine, a committee for the promotion of Information and Science Technology at MEXT (Ministry of Education, Culture, Sports, Science and Technology).
ここでコンテンツ終わりです。