Hacker News Reader

Comment by nickbauman | Hacker News Reader

There are currently no signs that what you think will happen will happen. Soft AI is the only place where anything is moving on that front and the movement is infinitesimally small. Here's an analogy for you: It took more than 1000 years (from Babylon to Archaic Greece) for us to go from writing with only consonants to using vowels for the first time.

View on HN

Years don't make progress on their own, people working during those years push progress forward. The estimated population of ancient Babylon at its height was 200,000. Let's imagine that 1% of them were working on developing writing for at leat 2 hours every week and that those who came after them were able to maintain that level of work for 1000 years until ancient Greece, over 200 million hours of work. That's less time than the official Gangnam Style video has been watched on YouTube.

In 50 years, 99%+ of all the work ever done by civilization will be done after 2016.

As long as we're criticizing the analogies in the discussion (rather than the actual arguments) I'd say the hours spent do not have a consistent quality vis-a-vis solving hard problems. Because there are more absolute hours available does not mean that there are more hours available for solving hard AI problems. There are very likely less. And there has been virtually NO progress on the hard AI front.

Analogies are always a rabbit hole. Haha.

Hard, human-level AI would help this a lot, but it isn't necessary. All that's required for traditional programming to become obsolete is for computers to be much better at understanding ambiguity and have a robust model for the flow of programs. With today's neural networks and technology, I have no doubt it would be possible to design something that would create good code based on all the samples on github. Not easy by any means or someone would have done it, but it doesn't require any breakthroughs of computer science, just lots of data and good design. The tools referenced in the articles are working primitive versions of this.

There's an important distinction though between being able to write a compiling (or even functional) program and being able to write a program that serves a particular purpose.

I'm talking about human-guided programming without using traditional programming language, creating a design document to lay out what it does and how data flows and allowing the computer to sort out the details based on a stored data set.

Why do you assume that this design document would be simpler to create than the traditional computer program? Because otherwise, this is exactly what happens now.

Design is always going to be a part of creating something. What this article is arguing is that manual typing of text by humans using traditional programming languages will not be the primary means of implementing those designs in the future. We don't yet know how to make computers into good designers, but we know that we can create tools that translate designs into executable code that can be less error-prone and more reliable than people typing letters into a text editor.

My question is how is drawing rather than writing simplifying anything i.e. what is the gain from moving from traditional programming to some sort of theoretical picture programming? Is it that you can draw lines between things rather than just assuming that the line from one symbol points to the next symbol on the line? Does that simplify things, or make them more complicated?

> we know that we can create tools that translate designs into executable code that can be less error-prone and more reliable than people typing letters into a text editor.

I disagree. Maybe you know, but I haven't seen any indication of the sort.

Drawing rather than writing is just one method. A lot of it will likely be conversational. I could imagine a designer with an AR overlay speaking to a computer which offers several prototypes based on an expressed intent. The designer chooses one and offers criticism just as a boss would review an alpha version and suggest changes. The machine responds to the suggestion and rewrites the program in a few fractions of a second. The designer continues the conversation, maybe draws out some designs with a pencil, describes a desire, references another program which the machine analyzes for inspiration, and the machine adjusts the code in response. This is just one of many possible examples. The point is that software design is trending toward more automation. Coding is not a new essential skill that everyone will need on the future. Human-machine interactions are trending toward natural and automated methods, not manual code entry. Most people need to learn to be creative, think critically, analyze problems, not learn the conventions of programming languages.

There are some fairly aspirational claims about how it might be different in this paper, which is a great read:

http://shaffner.us/cs/papers/tarpit.pdf

There has already been some significant progress on this front. E.g., SQL and logic programming let you describe what you want to happen, and let the computer figure out some of the details. Any compiler worth using does this, too. Smarter machines and smarter programs will mean smarter programming languages.

> I'm talking about human-guided programming without using traditional programming language, creating a design document to lay out what it does and how data flows and allowing the computer to sort out the details based on a stored data set.

Creating clear and accurate design documents is so much harder and more specialized a skill than programming that many places that do programming either avoid it entirely or make a pro-forma gesture (often after-the-fact) in its direction.

(I am only about half-kidding on the reasoning, and not at all about the effect.)

"creating a design document to lay out what it does and how data flows and allowing the computer to sort out the details based on a stored data set."

This is exactly what programmers do today. We just call the "design document" a "program".

Over time, our design documents become higher and higher level, with the programmer having to specify fewer details and leaving more of the work of sorting out the actual details to the computer.

Yes, exactly! That's what the article is claiming.

> for us to go from writing with only consonants to using vowels for the first time

Speaking as someone who has studied cuneiform and Akkadian, I would say that this claim isn't true. Here's a vowel that predates the period that you mentioned[0].

[0] https://en.wikipedia.org/wiki/A_(cuneiform)

> Here's an analogy for you: It took more than 1000 years (from Babylon to Archaic Greece) for us to go from writing with only consonants to using vowels for the first time.

Where did you get this idea? Babylonian writing fully indicated the vowels. It always has. You're thinking of Egyptian / Hebrew / Arabic writing.

Even where a semitic language was written in cuneiform, vowels were always indicated, because the cuneiform system didn't offer you the option to leave them out. https://en.wikipedia.org/wiki/Akkadian_language#Vowels

(Old Persian was written in repurposed cuneiform, and therefore could have omitted the vowels, but didn't.)

>It took more than 1000 years (from Babylon to Archaic Greece) for us to go from writing with only consonants to using vowels for the first time

Yeah and it took "us" 60 years from discovering flight to landing a rocket on the moon. Took "us" 60 years from the first computer to globally-live video streaming in your pocket. Time is a pointless metric when it comes to technology. You don't know what someone is cooking up down in some basement somewhere that will be released tomorrow and shatter your concept of reality.

I wonder if 'leisure-person years' is a better metric of progress (where 'leisure' is defined as the number of hours you can spend neither raising/searching for food nor sleeping).

Be really hard to identify, though.

There are signs, if you know were to look for them: http://arxiv.org/abs/1510.07211 https://arxiv.org/abs/1410.4615 http://arxiv.org/abs/1605.06640 https://arxiv.org/abs/1410.5401

Reply on news.ycombinator.com